investigation of potential reasons to account for …949713/fulltext01.pdf · construction, a wind...

INVESTIGATION OF POTENTIAL REASONS TO ACCOUNT FOR THE

UNDERPERFORMANCE OF AN OPERATIONAL WIND FARM

Dissertation in fulfillment of the requirements for the degree of

MASTER OF SCIENCE WITH A MAJOR IN ENERGY TECHNOLOGY WITH

FOCUS ON WIND POWER

Uppsala University

Department of Earth Sciences, Campus Gotland

RENAS TÜCER

27.05.2016

INVESTIGATION OF POTENTIAL REASONS TO ACCOUNT FOR THE

UNDERPERFORMANCE OF AN OPERATIONAL WIND FARM

Dissertation in partial fulfillment of the requirements for the degree of

MASTER OF SCIENCE WITH A MAJOR IN ENERGY TECHNOLOGY WITH

FOCUS ON WIND POWER

Uppsala University

Department of Earth Sciences, Campus Gotland

Approved by:

Supervisor, Simon-Philippe Breton

Examiner, Jens Nørkær Sørensen

27.05.2016

iii

ABSTRACT

Wind farms are costly projects and prior to the construction, comprehensive wind

resource assessment processes are carried out in order to predict the future energy yield

with a reliable accuracy. These estimations are made to constitute a basis for the

financial assessment of the project. However, predicting the future always

accommodates some uncertainties and sometimes these assessments might overestimate

the production. Many different factors might account for a discrepancy between the pre-

construction wind resource assessment and the operational production data. This thesis

investigates an underperforming wind farm in order to ascertain the reasons of a

discrepancy case. To investigate the case, the relevant data and information along with

the actual production data of three years are shared with the author. Prior to the

construction, a wind resource assessment was carried out by an independent wind

consultancy company and the work overestimated the annual energy production (AEP)

by 19.1% based on the average production value of available three years.

An extensive literature review is performed to identify the possible contributing causes

of the discrepancy. The data provided is investigated and a new wind resource

assessment is carried out. The underestimation of the wind farm losses are studied

extensively as a potential reason of the underperformance.

For the AEP estimations, WAsP in WindPro interface and WindSim are employed. The

use of WindSim led to about 2-2.5% less AEP estimations compared to the results of

WAsP. In order to evaluate the influence of long term correlations on the AEP

estimations, the climatology datasets are created using the two different reanalysis

datasets (MERRA and CFSR-E) as long term references. WindSim results based on the

climatology data obtained using the MERRA and CFSR-E datasets as long term

references overestimated the results by 10.9% and 8.2% respectively.

iv

ACKNOWLEDGEMENTS

I would like to take this opportunity to thank a few people who made this thesis possible.

First of all, I would like to thank my thesis supervisor Simon-Philippe Breton for his

valuable guidance and time which helped me a lot to shape my thesis.

I would like to thank Jens Nørkær Sørensen for his examination and critics to improve

my thesis work.

I would also like to thank the entire Wind Power Project Department at Uppsala

University Campus Gotland for their continuous support. I would like to thank Nikolaos

Simisiroglou for his friendly assistance using WindSim set-up.

I would like to thank the company which provided me with all needed data and

information to carry out this research.

I would like to express my deep gratitude to my parents. It would not have been possible

for me to be here without their unfailing support and encouragement throughout my life.

I would like to thank all my friends who colored my life with their presence. Finally, I

would like to thank Lorella Clausse for her patience and love which always encouraged

me to go forward.

v

NOMENCLATURE

AEP Annual Energy Production

CFD Computational Fluid Dynamics

CFSR Climate Forecast System Reanalysis

IEA International Energy Agency

IEC International Electrotechnical Commission

MCP Measure Correlate Predict

MERRA Modern Era Retrospective-Analysis for Research & Applications

NCAR National Center for Atmospheric Research

RANS Reynolds Averaged Navier Stokes

TI Turbulence Intensity

WAsP Wind Atlas Analysis and Application Program

WRA Wind Resource Assessment

vi

TABLE OF CONTENT

ABSTRACT ................................................................................................................... III

ACKNOWLEDGEMENTS .......................................................................................... IV

NOMENCLATURE ........................................................................................................ V

TABLE OF CONTENTS .............................................................................................. VI

LIST OF FIGURES ................................................................................................... VIII

LIST OF TABLES .................................................................................................... VIIII

CHAPTER 1. INTRODUCTION .................................................................................... 1

1.1 JUSTIFICATION OF THE RESEARCH ................................................................ 3

1.2 RESEARCH PROBLEM AND THE SCOPE OF THE RESEARCH....................... 3

1.3 THESIS OUTLINE ................................................................................................... 4

CHAPTER 2. LITERATURE REVIEW ....................................................................... 5

2.1 THESIS BACKGROUND ........................................................................................ 5

2.2 CONCLUSION OF THE BACKGROUND STUDIES AND THE NEED FOR A

NEW RESEARCH.......................................................................................................... 7

2.3 POTENTIAL REASONS OF THE DISCREPANCY .............................................. 7

2.3.1 DATA ................................................................................................................ 8

2.3.1.1 Topography

2.3.1.2 Wind Data

2.3.2 LONG TERM CORRELATION ..................................................................... 11

2.3.3 MODELLING .................................................................................................. 13

2.3.3.1 WAsP

2.3.3.2 WindSim

2.3.4 RELIABILITY OF MEASUREMENT SET-UP ............................................ 16

2.3.5 TECHNICAL PROBLEMS AND POWER CURVE DEVIATION .............. 17

2.3.6 LOSSES ........................................................................................................... 17

2.3.6.1 Wake Losses

2.3.6.2 Additional Losses

2.4 UNCERTAINTY ASSESSMENT .......................................................................... 24

2.4.1 WIND RESOURCE UNCERTAINTY ........................................................... 25

2.4.1.1 Sensor Accuracy

2.4.1.2 Sensor Calibration

2.4.1.3 Boom and Mounting Effects

2.4.1.4 Long Term Wind Prediction

2.4.1.5 Wind Flow Modelling

2.4.1.6 Other Uncertainties

2.4.2 ENERGY ASSESSMENT UNCERTAINTY ................................................. 27

2.4.2.1 Power Curve


vii

CHAPTER 3. METHODOLOGY AND DATA .......................................................... 28

3.1 DESCRIPTION OF THE CASE ............................................................................. 28

3.2 TOPOGRAPHY ...................................................................................................... 29

3.2.1 ELEVATION DATA ....................................................................................... 30

3.2.2 ROUGHNESS DATA ..................................................................................... 30

3.2.3 OBSTACLES ................................................................................................... 32

3.3 MONITORING EQUIPMENT ............................................................................... 33

3.4 WIND DATA .......................................................................................................... 34

3.4.1 DATA VALIDATION..................................................................................... 35

3.5 PRODUCTION DATA. .......................................................................................... 37

3.6 MEASURE CORRELATE PREDICT (MCP) ........................................................ 38

3.7 WIND FARM MODELLING TOOLS ................................................................... 40

3.7.1 WINDPRO ....................................................................................................... 41

3.7.2 WINDSIM ....................................................................................................... 42

3.7.2.1 Terrain Module

3.7.2.2 Wind Fields

3.7.2.3 Objects

3.7.2.4 Wind Resources

3.7.2.5 Energy

3.7.2.6 Grid Independency Study

3.8 LOSSES .................................................................................................................. 48

CHAPTER 4. APPLICATION OF THE METHODLOGY AND RESULTS ......... 50

4.1 INTERPRETATION OF WAKE LOSSES. ......................................................................... 50

4.1.1INFLUENCE OF WAKE MODELS ................................................................ 50

4.1.2 SECTOR-WISE WAKE LOSSES ................................................................... 52

4.2 SIMULATION RESULTS ..................................................................................... 53

4.3 UNCERTAINTY ASSESSMENT ......................................................................... 56

CHAPTER 5. DISCUSSION AND ANALYSIS .......................................................... 59

CHAPTER 6. CONCLUSION ...................................................................................... 62

REFERENCES ............................................................................................................... 65

APPENDIX ..................................................................................................................... 68

viii

LIST OF FIGURES

Figure 1: Schematic illustration of the process involved in the long term correlation

Source: Liléo et al. 2013 .................................................................................................. 13

Figure 2: The speed up effect as modelled by WAsP Source: Wallbank 2008............... 14 Figure 3: Flow separation due to the high slope angle Source: Nilsson 2010 ................. 15 Figure 4: Visualization of the wake formations behind the rotor according to the Jensen

Model on the left and Larsen model on the right Source: Renkema 2007 ....................... 20 Figure 5: Typical ranges for the additional losses. Source: Mortensen 2011 .................. 22

Figure 6: Visualization of the terrain in WindSim ........................................................... 30 Figure 7: Terrain Inclination (deg.) Source: WindSim .................................................... 31

Figure 8: Elevation grid data and roughness lines (blue lines) in WindPro (Centre square

represents the high resolution elevation data) Source: WindPro ..................................... 32 Figure 9: Monthly fluctuations of power production for three years ............................... 37 Figure 10: Horizontal grid resolution (left) and schematic view of the vertical grid

resolution (right) Source: WindSim ................................................................................. 43 Figure 11: Turbines and met masts located in the objects module Source: WindSim .... 45

Figure 12: Mean wind speed at 60m based on 3 climatology objects (grey points

represent the climatology object locations) Source: WindSim ........................................ 46 Figure 13: Changes of AEP for different cell numbers in WindSim for all turbines ....... 48

Figure 14: Energy yield and wake losses per sector Source: WindPro ............................ 53 Figure 16: Weighted mean of sector wise correlation between M1 and MERRA ........... 68

Figure 15: Tower Shading Flagging in Windographer .................................................... 68 Figure 17: Weighted mean of sector wise correlation between M2 and MERRA ........... 69

Figure 18: Weighted mean of sector wise correlation between M3 and MERRA ........... 69 Figure 19: Weighted mean of sector wise correlation between M1 and CFSR-E ........... 69

Figure 20: Weighted mean of sector wise correlation between M2 and CFSR-E ........... 70 Figure 21: Weighted mean of sector wise correlation between M3 and CFSR-E ........... 70 Figure 22: Climatology characteristics for M1/MERRA ................................................. 71

Figure 23: Climatology characteristics for M2/MERRA ................................................. 71 Figure 24: Climatology characteristics for M3/MERRA ................................................. 71

Figure 25: Climatology characteristics for M1/CFSR-E .................................................. 72 Figure 26: Climatology characteristics for M2/CFSR-E .................................................. 72

Figure 27: Climatology characteristics for M3/CFSR-E .................................................. 72

ix

LIST OF TABLES

Table 1: Long term correlation uncertainties corresponding to specific correlation rates

Source: GH and Partners 2009 ......................................................................................... 27

Table 2: Details of the set-up of equipment Source: the original WRA document......... 34 Table 3: The used period of time series and the coverage after the removal of invalid

data ................................................................................................................................... 36 Table 4: Discrepancy between the original AEP estimation and the operational data..... 38 Table 5: Details of long term reference datasets .............................................................. 39

Table 6: Grid spacing and the number of cells ................................................................. 43 Table 7: Distribution of first 10 nodes in z direction ....................................................... 43

Table 8: Details regarding to climatology data obtained by long term correlation with the

MERRA dataset Source: WindSim .................................................................................. 45 Table 9: Details regarding to climatology data obtained by long term correlation with the

CFSR-E dataset Source: WindSim ................................................................................... 45

Table 10: Assumptions of additional losses ..................................................................... 49 Table 11: Calculated wake losses ..................................................................................... 50

Table 12: Simulation results for all climatology data in WindSim and WindPro ............ 54 Table 13: AEP results with MERRA and CFSR-E datasets as long term reference ........ 55 Table 14: Discrepancy of the results compared to the operational data ........................... 56

Table 15: Assumed uncertainties associated with the WRA and AEP calculations ........ 58 Table 16: Probability of exceedence for different time intervals ..................................... 58

Table 17: Convergence table Source: WindSim .............................................................. 70

1

CHAPTER 1. INTRODUCTION

Energy is the yield of wind farms. Wind farms are designed and constructed as a result

of comprehensive projects which usually predict the expected energy yield with a

reliable accuracy and constitute a basis for the financial evaluation of the project. In the

process of developing a wind farm, in order to decide whether the investment is feasible

or not, detailed annual energy production (AEP) estimations are needed. A wind

resource assessment (WRA) is the key element for accurate AEP estimations and it

comprises a large process from the data acquisition to the handling and use of the data.

WRA is the assessment of micro wind climatology of the proposed site which is

performed in order to estimate the future energy production of the wind farm.

However, wind is not something easy to forecast and shows momentary, daily, seasonal,

and annual variations. A WRA procedure aims to make a reliable prediction of the future

wind for the wind farm’s life span and this work usually commences with the on-site

wind measurements. In order to avoid the seasonal bias, usually at least 12 months on-

site wind measurements are carried out. However, even though the data is collected

consistently, representativeness of the obtained data might be limited due to the

defective data and such data is needed to be detected. Moreover, even if the data

captured the wind characteristic of the entire year flawlessly, the wind might show big

annual variations. In order to reduce the uncertainty in the future winds, the data

acquired is usually correlated with a nearby long term dataset which contains about 20-

30 years of data. This correlation is named as long term correlation. Although this

process decreases the uncertainty for the future AEP estimations, it introduces a new

uncertainty into the estimations, called long term correlation uncertainty. This means

that the use of an improper long term reference carries the risk of creating even more

bias.

2

Moreover, wind farms are usually modelled with some wind farm modelling tools and

these models also can cause some biases. The use of an improper modelling tool and

wake model, underestimations of losses, or even hardware and set-up problems of

measurement instruments might account for a discrepancy between the predictions and

reality. All these factors add some uncertainty to the assessment and the uncertainty

might be decreased but not be completely mitigated. Different WRA approaches might

enhance or lessen the accuracy of pre-construction AEP estimations and might account

for a discrepancy between the estimated AEP and the operational outcome. A

discrepancy case might be the result of one of the abovementioned problems or a

combined consequence due to the accumulation and interaction of many problems.

This thesis investigates the reasons behind an existing underperformance case for a wind

farm located in a mid-complex terrain in the Republic of Ireland. Available operational

data displays an underperformance of production compared to previously made AEP

estimation. Reasons behind this discrepancy were unknown at this stage of the

investigation. A detailed WRA approach is carried out with the available data. The data

is handled and the representativeness of the data is evaluated. Two different wind farm

design tools namely, WindSim and WindPro, are employed and investigated for their

modelling accuracy. Available wake models in WindPro and WinSim are analyzed for

their impact on the results. Two different long term references (MERRA and CFSR-E)

are specified and the simulations have been carried out with both of them. This is done

in order to investigate the influence of long term references and to assess the success of

the correlations for reducing the discrepancy in the AEP estimations. Further researches

are performed in order to identify the potential reasons which might possibly account for

the discrepancy. The data provided, simulation results and the literature survey are the

main sources of the research and they are benefited together throughout the work. The

work identified the some of the reasons with their possible influence on the results while

some other possible reasons could not be numerically investigated due to the limited data

availability.

3

1.1 Justification of the research

Underperformance is not a rare problem encountered in the wind farm projects.

According to the consulting company DNV (2011), based on the wind farms available in

DNV database, 9% of all projects carried out in 2009 and 2010 displayed an

underperformance from the projected AEP estimation. There are many probable reasons

which might form a basis for an underperformance. As every underperformance case has

its unique conditions, identification of these reasons in this work aims to provide a

reference study for the future projects. This research investigates a specific operational

wind farm which is commissioned in the mid of 2010 and the production data of the

wind farm for 2013, 2014 and 2015 is available. Prior to the construction of the wind

farm, a WRA is carried out by an independent company and the results overestimated

the AEP. In this work, the case experiences up to 28.5% discrepancies (for 2014)

between the WRA and operational data. The discrepancy between the WRA and the

average production of three available years is 19.1%. An extensive research is needed to

address the contributory factors and to explain this discrepancy.

1.2 Research problem and the scope of the research

The main objective of this thesis is to identify the possible reasons contributing to the

discrepancy and to investigate their influence on the overall results with due diligence.

Through this approach, how these factors can be minimized or mitigated will be

discussed. The base for the research is the data provided by the company which currently

owns the wind farm. For this work, data and a great majority of details are confidential

therefore only required information will be shared with the reader. Data regarding to the

terrain, measurements, measurement instrument specifications, data calibration

certificates, wind turbine specifications and the actual production are provided along

with the original WRA document. However the data still limits the research since the

data acquisition is not carried out by the author of this thesis. The limitations of the study

are stated in Chapter 5. An extensive literature review is done to identify the possible

factors and all possible reasons are mentioned regardless from the focal point of the

4

thesis. These factors are divided into two categories, factors cannot be investigated due

to data limitation or time constraint and factors that can be investigated with the

provided data and within the given time course.

All provided data is analyzed and undergoes a handling process. Data is investigated

from various aspects with the help of available tools namely, WindPro, WindSim and

Windographer. Long term correlation is carried out and the potential impact of the long

term reference datasets is studied. Since the terrain is moderately complex, modelling

error is assessed by employing an analytical and a numerical wind flow modelling tool

namely, WAsP and WindSim. Furthermore, the wind farm is subjected to high wake

losses due to low distance between the wind turbines which sometimes can be as low as

2.8 rotor diameters. Therefore, simulations with all of the wake models available in the

employed tools are performed and their representativeness of the real situation is

discussed based upon the literature review and the results. The rest of the losses are

studied and the numerical estimations are made depending on the literature review and

the information provided by the company. Later on, the uncertainties involved in the

entire process are evaluated and applied to the results. Finally, the results obtained and

the actual operational data are compared. The improvements achieved are weighed by

the reduced discrepancy between the new assessment and the operational production

data. The eliminated discrepancy is analyzed and further research which is not covered

due to data limitations and time constraint is recommended.

1.3 Thesis outline

Chapter 2 studies the previous researches carried out to investigate the similar cases and

also acknowledges the potential justifications for the discrepancy and constitutes a base

for the methodology of the investigation. Chapter 3 introduces the case of the study to

the reader and explains the methodology followed in a detailed way. Chapter 4

comprises the results obtained by the application of the methodology and justifies the

results comparing them with results from the literature. Chapter 5 discusses the findings

5

and interprets the outcomes. Chapter 6 concludes the findings and suggests the points

that could not be investigated as a further research.

CHAPTER 2. LITERATURE REVIEW

The literature review chapter constitutes a basis for methodology chapter. The practices

followed in the methodology part are reasoned on the basis of the literature survey. In

this chapter, previous studies which constitute a background for this thesis are

introduced to the reader with their related findings. Similarities and differences are

discussed and the need for the present research is justified. Afterwards, a literature

survey has been carried out in order to learn the state of the art for the WRA and to

identify the potential reasons of the discrepancy observed. Identified factors are

classified as to whether it is possible to investigate them in this work or not. If an

investigation is possible with the available data and knowledge, a comprehensive

literature survey is made to form a valid methodology.

2.1 Thesis background

Previous works investigating the related subjects are needed to be studied to constitute a

background for the research. It is known that there are studies comparing the accuracy of

different wind flow modelling tools and wake models for different case studies and

topographical properties. WAsP and WindSim are two commonly used tools for wind

farm modelling therefore these two tools are frequently examined for their accuracy

under different conditions.

Berge et al. (2006) make a comparative study using WAsP and two other CFD based

tools namely, WindSim and 3DWind for a complex terrain. The emphasis of the work is

the evaluation of vertical and meso-scale variations in wind speed across the farm and

the variation of turbulence with the height. The study concludes that despite the

inapplicability of WAsP to the complex terrains, use of CFD based models did not lead

to an improvement of the average wind speed estimations. Teneler (2011) makes a study

6

to validate the use of WindSim as a CFD tool for a forested area and concludes that both

WAsP and WindSim overestimate the AEP for the studied case. However with the use of

forest model in WindSim, more accurate results are obtained. Simisiroglou (2012)

compares the AEP results employing two wind farm modelling tools namely, WindSim

and WindPro for a site located in Greece. The main focus of the work is to compare the

performance of the simulation tools for a specific case. Although the long term

correlation is crucial for the accurate AEP estimations, due to the absence of long term

reference data and poor correlation coefficients with the online datasets, long term

correlation is skipped. The investigation of the difference in AEP with the long term

correlation of wind speed measurements is rather suggested for further researches.

According to Simisiroglou (2012), different height contours, wake models and the

terrain type do not influence the AEP significantly in WindPro while roughness lines

have an apparent impact on the AEP for the simulations carried out. And the study

concludes that the percentile difference for two models (WindPro and WindSim)

estimating AEP with wake losses is in the vicinity of 1%.

Timander & Westerlund (2012) carry out a comparative study employing WindSim and

WindPro for a small wind farm located in the inland of Sweden featuring a fairly

complex and forested terrain. The work is resulted with the similar estimations from

both software and it is concluded that the site was not complex enough to show the

potential benefits of the CFD-based model (WindSim) for complex terrains. Mancebo

(2014) compares two commercial WRA tools namely, WindSim and Meteodyn WT for

their ability of predicting the vertical wind profile before and after an embankment site.

According to Mancebo (2014), Meteodyn WT predicts vertical wind profiles better

before the embarkment, while WindSim makes closer predictions just after the

embarkment. Finally, Gallagher (2014) investigates a discrepancy case for a moderately

complex site with the focus of inadequate wind flow modelling, employing WindPro and

WindSim for the comparison. The case of study experiences an average discrepancy of

%10.19 between WRA and operational data. Gallagher (2014) uses the MERRA dataset

7

as a long term reference and specifies the Jensen wake model as the appropriate wake

model for the case. And the study concludes that WindPro overestimates the AEP by

8.10% while WindSim underestimates the AEP just by 0.11 % therefore the reason of

the discrepancy is concluded to be the modelling of the wind farm using the WAsP

software out of its suggested envelope. Gallagher (2014) also suggests some further

work such as performing a more precise evaluation of the shadow and speed up effects

from the anemometers and the application of the uncertainties in the results.

2.2 Conclusion of the background studies and the need for a new research

It is now clarified that there are existing researches and studies for the comparison of the

accuracy of wind flow modelling tools which are also utilized in this thesis. However,

the focal point of the studied works is usually the performance evaluation of modelling

software. Therefore, they do not always include all the essential points for an accurate

WRA. For instance, Simisiroglou (2012) skips the long term correlation while any WRA

assessment without it would display a really high uncertainty. Gallagher (2014) also

investigates a discrepancy case however the work concludes that the use of WAsP is the

main reason explaining the discrepancy. For the case of the present study, the

discrepancy is too large (up to 28.5%) to be expected to solely be caused by the use of an

improper modelling tool. Therefore this case requires a broader investigation. This work

studies a large number of investigatable potential factors which might be possibly

contributing to the discrepancy within the delimitations of the work. Therefore this thesis

can be seen as a continuation of existing studies in order to explain an ample discrepancy

issue. On the other hand, each underperformance case might have its own specific

justifications therefore they are needed to be investigated unless the reason is obvious.

2.3 Potential reasons of the discrepancy

A literature survey regarding the potential factors that might account for the discrepancy

is carried out. Identified factors are investigated under the six subchapters namely, data,

long term correlation, modelling, reliability of measurement set-up, technical problems

and power curve deviation, and losses. It is concluded that the influence on the

8

discrepancy of data, long term correlation, modelling and, losses can be investigated

with the available data and knowledge. However the research-ability of technical

problems, power curve deviation, and measurement set-up is limited due to the absence

of the relative data or non-reproducible conditions. Finally, the uncertainties involved in

the entire process are studied at the end of this chapter.

2.3.1 Data

For this work, the data is the main source of analysis therefore data, data quality, data’s

impact on the results are needed to be discussed and investigated. Topography and wind

data are most essential data for the energy analysis. Without the presence of these

datasets, it is impossible to assess the micro wind climatology of the site.

2.3.1.1 Topography

Topography of the site consists of all ground properties present on the site and it is the

fusion of orography and surface roughness. Orography represents the changes in altitude

of the ground while roughness represents the surface properties. Topography as a whole

shapes the wind profile of the site and should be handled in a detailed way. Orography

might influence the wind in many different ways. It might cause an increase in wind

speed over a smooth and not so steep hill or if the orography is complex, it might induce

turbulence and decrease the wind speed (Simisiroglou 2012). All these ground properties

are represented as elevation, roughness and obstacle data in the simulations.

Also, the surface roughness and obstacles might have a great influence on the site’s wind

profile. According to Simisiroglou (2012), different roughness data used in the study

influenced the AEP results up to 7.3% in WindPro. Obstacles obstruct the wind and

influence the flow in surrounding area. According to Troen & Lundtang Petersen (1989),

a present obstacle might affect the wind profile vertically approximately three times the

height of the obstacle and between 30 and 40 times the height in the downstream

direction. Due to the great influence of obstacles on the flow, obstacles at a distance

9

from the site should be included in the simulations. For the minimum extent of the used

topography length, 100 times the height of present obstacles or 10km is widely cited as a

rule of thumb (Mortensen 2011). According to Mortensen et al. (2011), the domain size

should be preferably at least 5km extending from the wind site to any direction. These

guidelines are considered while choosing the domain size and properties of the

simulations. Concluding from the literature survey, it can be understood that the

digitalization of topography carries a great importance for accurate simulations. The

accuracy, resolution of the data and the domain size might have a role in the

discrepancy.

2.3.1.2 Wind Data

It might not be wrong to tell the wind data is the main source for the WRA. Wind data is

usually obtained through on-site met mast measurements. A met mast usually consists of

at least one anemometer and wind wane (usually several of them at different heights)

and preferably temperature and barometric pressure sensors. On-site measurements are

usually carried out for an entire year in order to avoid seasonal bias in estimations. Data

is logged by a data logger and the wind speed and direction are recorded with a pre-

determined averaging time interval. Time series usually contain the mean, max, min and

standard deviation values of wind speed and direction along with the ambient

temperature and pressure. Wind data might be a reason for the discrepancy by itself.

Sampling and the calibration of the data might accommodate some biases which might

contribute to the discrepancy. Also, erroneous data might arise from several reasons and

needed to be detected and mitigated.

Data Calibration

For the investigated project, the cup anemometers used for on-site measurements log the

data as Hz signals and these signal values cannot be directly used in the WRA.

Anemometers are calibrated in order to convert this signal data to wind speed data.

Anemometer calibration is made to define a relationship between output signal of the

10

anemometer and wind speed. Therefore, usually wind tunnel calibration tests according

to the well-accepted standards (e.g. MEASNET) are carried out individually for each

anemometer. In these tests, a wide range of wind speed is applied to the tested

anemometers and output signals are gathered in order to determine a transfer function

(slope and offset). Afterwards, this function is applied to the data thus the wind speed

data is obtained.

The transfer function (slope and offset) for cup anemometers either can be default

function previously established by testing a large number of sensors of the same model

or can be measured individually for the sensor that was purchased (Brower 2012).

However, one has to consider that the wind power is proportional to the cube of the wind

speed and small biases in wind data might cause much larger discrepancies. To acquire

such precision in wind data, it is recommended that individually calibrated anemometers

be employed (Coquilla et al. 2007). For this work, the manufacturer of the used

anemometers suggests to have an individual wind tunnel calibration (such as a

MEASNET calibration) performed on the anemometer/rotor pair and refers consensus

method (use of pre-established transfer function) as less accurate. For this study, the

anemometers used are calibrated individually as it is suggested by the manufacturer

however, even though data calibration is not expected to cause a considerable bias in the

WRA, there is still a calibration uncertainty which is needed to be taken into account.

Data Validation

Even though the wind data is recorded consistently, it might contain some defective data

which are not representative for the real conditions at the time of record. Such data

should be removed or replaced with another representative data if available. This process

is named as data validation and it is performed in order to decrease the uncertainties and

make data more representative. Icing, tower shadowing, flow obstructions, sensor

failures, improper mounting and some other reasons might make the data erroneous and

invalid for the assessment. Some software such as WindPro and Windographer are able

11

to handle the validation process by disabling the data which falls outside of the normal

ranges. Moreover, repetitive data and the data giving pre-set error value can also be

detected by these tools. Windographer is a industry leading tool which specializes on the

data handling. The software offers some useful tools such as icing and tower shading

detection. The tool detects faulty data with comparing several measurements at different

height to each other. Also, erroneous data can be detected by a flagging method which is

marking the data according to the specified parameters by user. In this work,

Windographer is used to identify tower shading while the icing is investigated by

checking the data manually.

Apart from purifying the data from faulty records, the observed wind climate should be

also as representative as possible. Since the on-site measurements are the main source of

long term production estimation, measured wind data should cover at least a one year

period. Furthermore, in order to avoid seasonal bias, an integer numbers of full years

should be used (Mortensen 2011). Also Liléo et al. (2013) suggests the use of complete

years of measurements. To set an example, if the measurement is covering January 1st

2007 to February 1st 2008, data from January 1

st 2008 to February 1

st 2008 should not be

taken into account.

2.3.2 Long term correlation

On-site wind measurements are usually carried out for a year or few years. Making long

on-site measurements to capture the micro climatology of the site is usually not possible

and feasible. However, wind might show big variations from year to year therefore

relying solely on several months of on-site measurement is quite risky for the WRA.

Annual changes in wind speed constitute a large uncertainty for wind power projects.

According to Troen & Lundtang Petersen (1989), up to 30% variations in wind resources

can be expected from one decade to another. In order to decrease this uncertainty in

WRA, correlation of the measured datasets with a long term reference data is needed.

Measure-Correlate-Predict (MCP) is a widely used technique for long term correlation.

12

To be able to apply this technique, at least one on-site measurement and a nearby long

term reference are needed. MCP technique uses the past wind records to predict the

future wind and it assumes that the long term wind climatology is stable therefore past

data can be used to predict the future. A long term reference data should be chosen

carefully. It should be consistent and representative of the site’s climatology.

Furthermore, long-term data should describe real changes in the local climatic

conditions, and should not be affected by artificial changes (Liléo et al. 2013). Distance

of the long term reference from the site, record interval and the data coverage are also

important parameters to achieve an accurate long term correlation. In climatological

practice a 30-year period is often taken as the basis (Troen & Lundtang Petersen 1989).

Therefore long term dataset should be also long enough to represent the climatology of

the area. On the other hand, regional wind trends are needed to be considered while

choosing the long term reference period. According to the Liléo et al. (2013), the period

of 1989-1995 was characterized by unusual high annual mean wind speeds in northern

Europe therefore use of these years are suggested to be avoided. Also in the original

WRA document, unusual high wind speeds for the region in the early 90’s are mentioned

and this period is not taken into account. Therefore use of this period is also avoided in

this work.

Reliability of the long term correlation depends on the many factors and usually assessed

with the correlation coefficient which indicates that how well is the compatibleness of an

on-site measurement with the long term data. Long term correlation is made to reduce

the future wind uncertainty however it creates long term correlation uncertainty. This

uncertainty is explained in details in the uncertainty subchapter. Long term correlation is

an important determinant for the accuracy of the WRA and might be accounting for the

discrepancy. Schematic illustration of the long term correlation process can be seen in

Figure 1.

http://tureng.com/tr/turkce-ingilizce/compatibleness

13

Figure 1: Schematic illustration of the process involved in the long term correlation

Source: Liléo et al. 2013

2.3.3 Modelling

There are many variables affecting the energy yield of the wind farms. Usually,

specifically developed wind farm design tools are employed to model the farm. There

are quite a few commercial tools such as WindPro, WAsP , WindSim and WindFarmer.

In this study WAsP in WindPro interface and WindSim are used to identify the influence

of different flow modelling approaches on the AEP estimations. There are several

reasons to employ these two tools. First of all, WAsP is used for the original WRA

therefore use of WAsP with an enhanced WRA approach enables us to evaluate the

impact of changed parameters. Use of WindSim as a CFD based numerical model,

allows us to compare the results of two different flow model approaches namely,

analytical model (WAsP) and numerical model (WindSim).

14

2.3.3.1 WAsP

WAsP is a linear model which is widely used and well accepted tool for commercial use.

It has been developed by Risø DTU National Laboratories. The wind statistics are

created from the climatology and topography data and the WAsP makes its prediction

based on the linear extrapolation of the wind statistics. In this study, WAsP is employed

in the WindPro interface. For micro-scale flow (spatial scales of 1 m to 2 km), the WAsP

model is most commonly used in the wind resource analysis, but in areas with flow

separation the model is not very well suited for resource assessment (Berge et al. 2006).

This is mainly because WAsP flow model does not handle the turbulence generated by

the orographic features such as steep hills. However, turbine induced turbulence is taken

into account by employing the available wake models. WAsP models the flow as

attached flow to the ground. This approach achieves a quite decent representation of the

real behavior of the flow for the low-complex terrains. When the wind is flowing over a

smooth hill, a speed-up effect is encountered. The speed-up effect modelled by WAsP

can be seen in Figure 2.

Figure 2: The speed up effect as modelled by WAsP Source: Wallbank 2008

15

However, when the terrain slopes exceeds a certain threshold (most commonly given as

approx. 17 degrees), flow separations occur and attached flow approach leads to an

overestimation of the hill-shape speed-up effects (Wallbank 2008) .

Figure 3: Flow separation due to the high slope angle Source: Nilsson 2010

Therefore use of WAsP in complex terrain might lead to an overestimation of the AEP

depending on the complexity of the site. In order to evaluate the accuracy of WAsP for

this case, WindSim is employed.

2.3.3.2 WindSim

WindSim is a CFD based numerical model. CFD is employed to solve fundamental fluid

flow equations in the wind fields module. WindSim applies a 3D wind field modelling to

calculate the wind characteristics. The terrain and the atmosphere above it are perceived

as a complete volume and the horizontal and vertical gridding is performed for the entire

volume. Instead of extrapolating the wind flow linearly from the measurement points to

the hub-height, WindSim solves RANS equations for every node on the defined grid

(horizontal and vertical grid). The Navier-Stokes equations are non-linear partial

differential equations known to be unstable and difficult to solve (Meissner 2010).

Therefore in WindSim, some simplified methods to solve the some troublesome non-

linear terms are developed. WindSim contains six modules namely, terrains, wind fields,

16

objects, results, wind resources and energy. Functions of these modules will be

explained in following chapters along with the application of the methodology.

According to Nilsson (2010), WAsP and WindSim estimate the energy production at a

similar level which is close to the measured production when the orographic complexity

is low. Unlike WAsP, the non- linear WindSim model can also handle the flow

separation on the terrain due to its turbulence models. Therefore WindSim is inclined to

estimate lower and more realistic AEP estimations for the complex terrains. WindSim

lacks the functions to digitalize the terrain therefore some other tools to digitalize the

data might be needed. Since the original WRA is made using WAsP, the influence of

two different flow models (WAsP and WindSim) on the discrepancy is needed to be

investigated.

2.3.4 Reliability of measurement set-up

Used sensor models and types and even positioning of measurement sensors might affect

the wind data. On-site measurement is a complex process and the problems might occur

in many steps. To set several examples, installation of the met-mast and measurement

units might not be properly done or the location of the met-mast might be experiencing

obstructed wind flow. Moreover, if sensors are not properly mounted or unintentionally

tilted, the data might become invalid. All these factors introduce some uncertainty to the

measurements however, following the international standards and guidelines such as

IEC, should limit the measurement uncertainty to acceptable levels.

As a guideline, if sensors must be placed near an obstruction, they should be located at a

horizontal distance of no less than 20 times the height of obstruction in the prevailing

wind direction (Brower 2012). Guidelines for boom lengths for lattice and tubular towers

are needed to be followed in order to avoid tower shading effect. Also, boom directions

are needed to be paid attention for accurate measurements. According to Mortensen

(2011), the optimum boom direction is at an angle of 90° (lattice mast) or 45° (tubular

mast) to the prevailing wind direction. Moreover, if all of the anemometers are mounted

17

with different boom directions along the mast, the tower effect on the speed readings

would be different for all of the anemometers. In order to make accurate estimations of

wind shear, having booms at different levels pointing the same direction is strongly

recommended (Brower 2012).

There might be slight deviations in the wind speed reading for different anemometer

models. Some anemometer uncertainties might be mitigated by using several types of

anemometers together. For instance, cup anemometer might be affected from vertical

wind speeds. While sonic anemometers are able to measure the vertical wind speed thus

the horizontal wind speed measurements are not affected by vertical winds. This issue is

taken into account while assessing the uncertainties of cup anemometers. Measurement

uncertainties might be contributing to the discrepancy. And eliminating uncertainties

increase the quality of the WRA.

2.3.5 Technical problems and power curve deviation

Wind turbines might be producing less due to some technical problems or wind turbines

might be experiencing lower power curves than the ones provided by the manufacturer.

Power curves are measured under the standardized conditions however wind properties

are unique to the sites depending on the terrain features. High or low shear and

turbulence intensity or the inflow angle might cause deviations in the turbine

performance. Therefore a deviation from the proposed power curve is expected for every

project. Poulos (2015) investigated 50 underperforming wind farms and concluded that

the large power curve underperformance is one of the main reasons for the large

discrepancies. However, no relevant data or report is available regarding the occurrence

of these problems in the studied farm, therefore the risk of large power curve deviation is

not investigated in this work.

2.3.6 Losses

Losses are encountered in the every steps of wind power generation. Underestimation of

losses might be an important factor accounting for the discrepancy. The losses evaluated

18

in this chapter are the wake, availability, electrical, environmental, curtailment and

turbine performance losses. Some losses are estimated through analytical models such as

wake losses while some others are estimated on the basis of a literature review. A

tendency to underestimate losses is one of the reasons why wind plants have often not

produced as much energy as predicted in pre-construction studies (Brower 2012).

Therefore the losses for the specific case of this study will be examined thoroughly.

In this section, losses are investigated under two captions. Wake losses which are

directly obtained through simulations and the additional losses estimated by studying

original WRA document and making additional literature survey. Data taken from

original WRA document is not cited to a document by the reason of confidentiality.

Wake losses vary depending on the used wake model and additional losses might show a

big variation but they are usually about 5-10% (Mortensen 2011).

2.3.6.1 Wake Losses

Wind turbines after extracting energy from the wind, in addition to making the

downstream wind turbulent, slow it down, which cause less energy yield for the

impacted turbines on the downwind direction. This effect is named as wake effect and

might lead to significant reduced energy yields as well as fatigue loads in the wind farm.

The corresponding losses are named as wake losses. Usually, the proximity of wind

turbines (measured by rotor diameters) has a great influence on the wake losses. The

wake formation occurs behind the rotors of turbines and the flow recovers after a

distance. In the case study of Neustadter & Spera (1985) , wake losses caused 10% of

power reduction for 3 wind turbines in a row with 7D distance to each other and the

wake can persist even longer if the turbulence is low, such as offshore applications

(Diamond & Crivella 2011). For the case of this study, distance between the turbines

sometimes goes as low as 2.8D so the farm is subjected to significant wake losses. In

this study, wake losses are calculated employing the available wake models in WindPro

and WindSim. WindSim contains three wake models namely, Jensen, Larsen and

19

Ishihara wake models while WindPro contains three wake models (Jensen, Larsen and

Eddy-Viscosity) and additionally two modified (Jensen 2005 and Larsen 2008) wake

models. The wake decay constant is used to specify how the wake is widening , with the

downstream distance from rotor and a default value of 0.075 is given in WindPro

(Timander & Westerlund 2012). Due to the nature of wake losses, the accuracy of wake

models is quite case dependant therefore the proper wake model should be chosen

specifically for the individual cases. Available wake models in WindPro and WindSim

are presented here to comprehend their suitability to the case of this study. The results

from previous studies regarding the wake models are surveyed in this chapter and these

findings will be compared to the results obtained in Chapter 4. The analytical wake

models investigated here directly or indirectly take several parameters into account such

as incoming wind speed (m/s), rotor diameter (m), downstream and radial distance from

the rotor (m) and, ambient turbulence intensity to calculate the velocity deficit and wake

widening.

Jensen Wake Model

The Jensen wake model assumes a linearly expanding wake with a velocity deficit that is

only dependent of the distance behind the rotor (Renkema 2007). The rate of wake

expansion depends on the wake decay constant therefore this constant influence the

results significantly. Turbulence intensity is not directly taken into account but a relation

between the wake decay constant and turbulence intensity can be established. Low

turbulence intensity means a low wake decay constant thus higher wake losses (Sørensen

et al. 2008). According to Renkema (2007), the Jensen wake model is a fine-tuned model

with the use of WAsP. Moreover, in the case studies of Sørensen et al. (2008), Jensen

model with a proper choice of wake decay coefficient is found to be performing better

than or as good as other available wake models in WindPro. Also, WindPro suggests the

use of the Jensen wake model.

20

Larsen Wake Model

The Larsen model is a semi-analytical wake model based on the Prandtl turbulent

boundary equations and has closed form solutions for the width of wake and the mean

velocity profile in the wake (Renkema 2007). Therefore Larsen model is not affected by

the wake decay constant. The velocity deficit is dependent on both the downstream and

radial distance from the rotor. Larsen model takes the ambient turbulence intensity into

account along with the thrust coefficient and the undisturbed wind speed. According to

Renkema (2007), Larsen model in WindPro displays an unsatisfying performance due to

too wide and shallow wake estimations. There are two available version of Larsen

model in WindPro (Larsen 1999 and Larsen 2008). However, the difference between

two versions is only relevant for the near-wake (zone directly behind the rotor)

(Sørensen et al. 2008) therefore the use of these two different versions is usually

expected to lead to the same results.

Figure 4: Visualization of the wake formations behind the rotor according to the Jensen

Model on the left and Larsen model on the right Source: Renkema 2007

N.O Jensen 2005 Wake Model

N.O. Jensen 2005 model is a modified version of the traditional N.O Jensen wake model

implemented by EMD. The main difference between the Jensen model and the modified

21

Jensen model is experienced when the wakes from the different turbines overlap. In this

case, the original Jensen model leads to increased wake losses while the modified

version gives results with relatively constant power losses (Renkema 2007). According

to Sørensen et al. (2008), the original (old) Jensen model is inclined to result in the

larger wake loss (lower park efficiency), which usually tends to be closer to the observed

wake loss.

Eddy Viscosity (Ainslie) Wake Model

The Eddy Viscosity model is one of the available models in WindPro. Eddy Viscosity

Model estimates the wake velocity decay using the time averaged Navier Stokes

equation and the eddy-viscosity approach (Timander & Westerlund 2012). In the case

studies of Sørensen et al. (2008), Eddy Viscosity model generally underestimated the

wake losses in WindPro. According to GH and Partners (2009), if the turbines are

closely spaced (around 2D) in a row, Eddy-Viscosity model is tended to underestimate

the wake losses for the subsequent downwind turbines.

Ishihara Wake Model

The Ishihara model is an analytical wake model which is available in WindSim. One

important parameter of this wake model is that it accounts for the effects of turbulence

on the wake recovery from both the ambient turbulence and the turbine-generated

turbulence (Tong et al. 2012). According to Charhouni et al. (2015), the wake recovery

is rather dependent on the turbine induced turbulence for this model.

2.3.6.2 Additional Losses

On the other hand, there are other losses such as availability, curtailment and

environmental losses which are not easy to model or predict therefore usually typical

values are benefited. Typical ranges of the losses can be seen in Figure 5.

22

Figure 5: Typical ranges for the additional losses. Source: Mortensen 2011

In the original WRA document, some losses are estimated through the statistical analysis

of the local grid and wind farms. Using loss assumptions originating from real regional

statistics are anticipated to be more accurate than using typical ranges obtained through a

literature survey. Therefore these values are assumed to be enforceable for this paper as

well and the unvalued losses are estimated benefiting the typical ranges of losses.

Availability

In the original WRA document, the turbine availability is assumed to be 97% based on

the data from modern operational wind farms. The grid availability is estimated as

98.3% and the substation maintenance loss is assumed to be 0.2% based on the data

provided by the developer. These numbers show parallelism with the typical ranges of

losses therefore these numbers are assumed to be true for this study as well.

Electrical efficiency

Some percentile of the generated electricity is lost due the transmission losses and the

consumption of the farm. Operational electrical efficiency of the farm is provided as

23

99.4% by the developer and the wind farm consumption loss is not specified. According

to Mortensen (2011), total electrical losses range from 1 to 2 percent.

Environmental losses

Ice and dirt accumulation, and other environmental effects may change the aerodynamic

pattern of the blades and reduce the energy generation. In the original WRA document,

environmental losses are estimated as 0.5% considering the local climate of the site.

Turbine performance losses

Wind farm performance might also be affected from the high-wind hysteresis and the

power curve deviation. The operational power curve is expected to deviate slightly from

the manufacturer’s power curve obtained under standard test conditions. Also, when the

turbine cuts-in or cuts-out due to the high or low wind speed, the operation of the

turbines does not restart until the wind speed is measured in the turbine’s operation

range for a while. Losses occurred during this cut-in and cut-out process is named as

high-wind hysteresis losses. The loss due to high-wind hysteresis is estimated as 0.6% in

the original WRA. However the power curve adjustment losses are not considered.

According to Mortensen (2011),the total turbine performance losses are usually in the

range of 1-2%.

Curtailment losses

Lastly, curtailment losses comprise several specific losses. Wind sector management,

power purchase agreement, flicker & shadow, noise and, bird & bat deaths might cause

the shutting-down of some or all of the wind turbines within the farm. As might be

expected, specifying typical ranges for the losses occurred due to such factors like

flicker & shadow or bird & bat deaths is not possible. Thus, the curtailment losses are

design dependant and vary for every project.

In the case of close turbine spacing, if the wind flows from a direction which is facing

the turbines as an aligned row, wind conditions within the site might require to shut

24

down some of the turbines. This mainly occurs due to the increased wake losses within

the wind farm and its influence on the wind quality. This loss is named as wind sector

management loss. According to Brower (2012), if the inter-spatial distance between the

turbines is less than 3D, a wind farm might be subjected to considerable curtailment

losses due to the wind sector management. According to Poulos (2015), large magnitude

wind farm underperformance is mainly driven by underestimated curtailment, low plant

total availability and large power curve underperformance. Unfortunately with the

available data, it is not possible to investigate the influence of curtailment losses on the

discrepancy. This loss is neither taken into consideration in the original WRA nor is

specific data provided regarding to it. Assessment of curtailment losses is not made.

However, the influence of this loss on the AEP might be another research subject.

2.4 Uncertainty Assessment

Wind as the fuel of wind farms is a variable and uncertain resource. The future wind can

only be predicted and predictions always accommodate some uncertainties. Wind

resource assessments are carried out to understand the prevailing wind climatology over

a site. Micro wind climate analysis is the basis for the feasibility assessment, choice of

spatial array, preference of turbine model and type, and many other important

parameters of wind farms. Accurate data plays a key role for a decent WRA. However,

no tower perfectly represents the entire area, no sensor measures with perfect accuracy

and no data gathered in a limited time period perfectly reflects conditions a wind plant

may experience during its life time (Brower 2012). All these biases contribute to the

uncertainty in the wind resource analysis. The uncertainty assessment is basically

defining the limits of possible errors which are likely to occur in the WRA. By defining

these limits, occurrence probability of different levels of the AEP can be estimated. P90,

P75 and P50 values stand for the exceedance probability of 90%, 75% and 50%

respectively (Mortensen 2011). To put it more explicitly, annual energy yield of the farm

is expected to be more than the P90 value with a 90% possibility. Thus credit institutions

usually utilize P90 value for their financial assessments. The WRA and modelling of the

25

wind farm aim to minimize the uncertainties so there would be not much difference

between P50 and P90 values. Mitigation of the uncertainties is not possible however

following best practices in data acquiring and WRA process limit the uncertainties

considerably. The main sources of uncertainty can be divided into two groups namely,

wind resource uncertainty and energy assessment uncertainty (Lira et al. 2014). In this

section, sources of uncertainty are introduced to the reader and in Chapter 4.3,

assumptions regarding the uncertainties is given in Table 15.

2.4.1 Wind Resource Uncertainty

Wind resource uncertainty refers to the uncertainties associated with the data acquisition

and the data handling process. It is mainly influenced by the sensor accuracy, sensor

calibration, boom and mounting effects, long-term wind prediction uncertainty and the

uncertainty in the wind flow modelling.

2.4.1.1 Sensor accuracy

Used sensors have an influence on the quality of record therefore well-established

sensors should be employed to decrease the uncertainty in measurements. Lira et al.

(2014) suggests the use of The First Class or Vector anemometers. For the on-site

measurements of the case study, Vector anemometers are employed. Also, it may be

desirable to deploy more than one model on each mast (Brower 2012).This approach

would decrease the risk of faulty data caused by a problem affecting an anemometer

model. Sensor accuracy suggestions are usually provided by the manufacturer. This

accuracy error margins assume that sensor is mounted perfectly.

2.4.1.2 Sensor calibration

Calibration of the sensors also contributes to the uncertainty. However the use of

standardized wind tunnel calibration tests such as MEASNET decreases the uncertainty

considerably. According to Lira et al. (2014), institutions which make calibration of the

anemometers usually guarantee the results to not accommodate more than a 0.5% bias.

26

2.4.1.3 Boom and mounting effects

Improper mounting and proximity between the sensor and tower might influence the

measurements. If the anemometer is too close to the tower, distortion uncertainty might

arise or if the axis of anemometer is not completely vertical, another uncertainty due to

the flow inclination might occur. In order to decrease the uncertainties associated with

boom and mounting, IEA’s standard guidelines should be followed. The IEA’s

guidelines recommends 0.5% error for tubular and lattice towers (Lira et al. 2014).

2.4.1.4 Long term wind prediction

Long term correlation decreases the uncertainty caused by year to year wind variability

in long term. Nevertheless, the long term correlation also introduces an uncertainty to the

estimations. This uncertainty arises mainly from predicting the future wind conditions

depending on the past records, representativeness of the reference data for the site’s

climatology and the correlation success. The error in the estimation of the long-term

corrected wind speed is about 1.5 to 2 % (Liléo et al. 2013). Also, the correlation

between the on-site measurements and long term reference data accommodates an

uncertainty since the extension of on-site measurement is achieved using the non-linear

transfer functions obtained through the correlation of parallel measured period of on-site

measurement and the reference dataset. Correlation coefficient depends on the

compatibleness of these two datasets for the parallel measured period. Even though there

is not a simple relation between the correlation coefficient and correlation uncertainty,

uncertainty assumptions corresponding to the correlation rates estimated by GH and

Partners (2009) can be seen in Table 1.

27

Table 1: Long term correlation uncertainties corresponding to specific correlation rates

Source: GH and Partners 2009

The total uncertainty of the long term correlation is a combined value of the mentioned

specific uncertainties. According to Liléo et al. (2013), the total uncertainty associated

with the long term correlation of 1 year on-site wind measurements is about 2.1 to 4.5%

and if the on-site measurement period is extended to 2 years, the uncertainty decreases to

about 1.8 to 3.6%.

2.4.1.5 Wind flow modelling

Even though, the wind flow models usually give highly accurate results, they have a

limit for simulating the real conditions. According to (Lira et al. 2014), a typical range

for wind flow modelling uncertainty is 3-6%. This uncertainty can be reduced by

employing proper wind flow models. For instance, use of WAsP in complex terrains

might cause larger biases which might be larger than about 5% (Mortensen 2011).


Apart from the mentioned reasons, almost every step of WRA accommodates some

uncertainties. For instance, wake losses are modelled by wake models and these models’

representativeness of real wake losses carries an uncertainty. Also, if not considered in

the modelling uncertainty, vertical wind speed extrapolation from measurement height to

hub height causes an uncertainty.

2.4.2 Energy Assessment Uncertainty

As energy yield has an exponential relation with wind speed, wind resource uncertainty

influences the uncertainty in energy yield significantly. Also there are uncertainties

28

occur directly in the phase of energy production. These uncertainties are named as

energy assessment uncertainty.

2.4.2.1 Power Curve

Power curve analysis of wind turbines are carried out under standardized test conditions.

However wind flow shows different features on different sites therefore power output

might not be the same with the ones provided by the manufacturer. Typical range of this

uncertainty is usually between 4-6% (Lira et al. 2014)


It is known that availability, transmission, curtailment and environmental losses occur in

the wind power production. These losses are estimated and these estimations

accommodate uncertainties.

CHAPTER 3. METHODOLOGY AND DATA

In this chapter, data and the followed methodology are introduced to the reader. Some

details regarding the previously made WRA are explained and evaluated and the

methodology is justified with the previously made literature review.

3.1 Description of the case

In this work, the investigated wind farm is named as Wind Farm A and is located in

Northern Ireland (UK). The site can be assessed as a mid-complex site. The farm

consists of eight wind turbines, each with 2.5MW rated power, 80m rotor diameter and

60m hub height. Site A is adjacent to an old wind farm consisting twenty wind turbines

with 45m hub height and 47m rotor diameter. The operation of Wind Farm A started in

summer 2010. In some parts of the farm, proximity between turbines can be as little as

2.8D therefore Wind Farm A is subjected to considerable wake losses. The wind farm is

also in the vicinity of a commercial forestry area. In order to investigate the wind farm,

all confidential available data is shared with the author. All of the provided data is

studied and inspected along with a literature review. After inspection of the data, data is

29

used for simulation with following the best practices learned through the literature

survey. The available software used for the energy calculations are WAsP 11 (Version

11.04.0006), WindPRO (Version 3.0) and WindSim 7.0.0.

Provided information contains high resolution elevation data, time series wind data,

measurement instrument specifications, calibration certificates, specifications of the used

turbine, site map and location information, production data and the needed information

regarding the adjacent wind farm. In this section, relevant information regarding the data

is introduced to the reader and the methodology followed is explained.

3.2 Topography

The wind farm is located on a small irregularly shaped hill with a maximum elevation of

approximately 300m and surrounding area is a flat plain with an average height of 100m.

Ground cover of the site contains grassland, swamp and forestry areas. A commercial

forestry area is present between 1.5 and 2km away from the turbines. Topography

properties are introduced to the simulations as elevation and roughness data. WindSim

can visualize the terrain using the elevation and roughness contours. The terrain and the

middlemost hill where the wind farm takes place can be seen in Figure 6.

30

Figure 6: Visualization of the terrain in WindSim

Present slopes of the site are visualized by the terrain inclination angles. The figure

created by WindSim to illustrate the inclination angles might be helpful to understand

the complexity of the site. As can be seen from Figure 7, inclination angle of the hill

varies from 5 to 30 degrees while the surrounding plain has a less complex profile. Flow

separations are expected to occur since steeper slopes than widely cited 17 degrees are

present on the terrain.

31

Figure 7: Terrain Inclination (deg.) Source: WindSim

3.2.1 Elevation Data

In order to simulate the case, elevation and roughness data are needed. Provided

elevation grid data contains 3x3km area and only covers the immediate vicinity of the

wind farm therefore the domain is extended to 20x20km by annexing online digital

topographic data available in WindPro (SRTM) database. This domain size is assessed to

be a proper size since the rule of thumb (10km from each dimension) is met. The

resolution of the elevation data is 1x1m for the centre area (3x3km) where the wind farm

is located. This resolution is the highest resolution that WindSim accepts. The online

elevation data for the surrounding area has a resolution of 36x31m.

3.2.2 Roughness Data

Roughness data are obtained through WindPro database from the digitized roughness

model of Corine land cover 2006. This data is provided to WindPro by the European

32

Environment Agency (EEA) and obtained through the visual interpretation of high

resolution satellite imagery. This model currently contains the most up-to-date and

highest resolution data in WindPro. According to Mortensen (2011), approx. 25x25km2

roughness map should be used for the wind turbines with 80 meters hub height. For the

Wind Farm A, due to the absence of any remarkable obstacle, 20x20km2 digitized

roughness map is used. The forestry area is needed to be incorporated into the

simulations. Forests can either be introduced as roughness contours or obstacles to the

WindPro simulation. In this work, forested areas are introduced as roughness contours.

Surface roughness length represents the height where wind speed is theoretically equal

to zero. Usually, roughness length for the forests and woodlands are taken as 0.4m to

1.2m and commonly a roughness class 3 or roughness length 0.4m is assigned for pine

forests (Wallbank 2008). A roughness class of 3 and 3.2 and, a roughness length of 0.9m

are assigned to the nearby commercial forest by Corine land cover database. Elevation

grid data and roughness lines are merged in WindPro and can be seen Figure 8.

Figure 8: Elevation grid data and roughness lines (blue lines) in WindPro (Centre square

represents the high resolution elevation data) Source: WindPro

33

Wind profile over an open terrain differs from wind profile over a forest. In order to

account for this difference, displacement height might be introduced to the simulations.

To prevent confusion, displacement height and roughness length are different

parameters. Increasing the forest height might not necessarily increase the roughness

length of the terrain but instead its displacement due to the lifted wind profile (Timander

& Westerlund 2012). For the case of this study, there is a small neighbouring forest but

turbines are located on an open terrain therefore no displacement height is applied.

3.2.3 Obstacles

Mortensen (2011) suggest that if the obstacle height is three times the hub height and the

obstacle is closer to the site than 50 times its height, it should be treated as a sheltering

obstacle otherwise it should be assessed as roughness. Since the height of forest is much

lower than the turbine hub height, commercial forestry area is attached to the map as

roughness data. Any other remarkable obstacle in the vicinity of wind farm is not

detected therefore no sheltering obstacle is added to the terrain.

3.3 Monitoring Equipment

There are three on-site measurement masts present namely, M1, M2 and, M3 and the

records from one distant measurement masts (M4) is available. M4 is approximately

17km from the site while M1, M2 and M3 are scattered on the site. Data is sampled as

10 minutes mean wind speed and wind direction for all measurements. Also min, max

values and, standard deviation of wind speed and direction are recorded. All of the

anemometers are individually calibrated according to MEASNET standards. The top

anemometers of M1 and M2 are mounted on vertical T-booms which extend from the

57m high met mast. M3 has two anemometers at the top height. M4 contains one side-

mounted anemometer and one side-mounted wind wane. All anemometers are mounted

on the booms approximately 10 mast diameters long and all anemometer cups are

mounted 15 boom diameter above the boom. Mounting arrangement is consistent with

34

the IEC standards. Campbell Scientific CR510 data loggers are used for the record of all

wind data and the details regarding to set-up of equipments can be seen in Table 2.

Table 2: Details of the set-up of equipment Source: the original WRA document

Instrument Height (m) Boom orientation (°)

Mast M1

Vector A100L2 Anemometer 60.1 330

W 200 P Wind vane 60.1 140


W 200 P Wind vane 53.5 218



Mast M2


W 200 P Wind vane 60.1 145


W 200 P Wind vane 53.5 218



Mast M3

Risoe P2546A Anemometer 60.1 135


W 200 P Wind vane 57.1 182

NRG Systems NRG 200P 56.9 272


NRG Systems NRG 200P 53.85 272


Mast M4


W 200 P Wind vane 10.6 247

3.4 Wind Data

M1 and M2 datasets cover slightly more than 12 months and they are available from

2007/02/12 and 2007/02/14 to 2008/03/10. M3 data set covers approx. 16 months from

2009/06/09 to 2010/10/12. Data from M4 contains 7 years of data with some gaps.

Dataset from M4 is too short to use as a long term reference and consists 4 months of

35

missing or suspect data in the time series. Due to its distance from the site and the

insufficient knowledge about the surrounding surface properties, M4 dataset is

disregarded in the simulations. In the original WRA, the M4 dataset was extended to 12

years by correlating it with the monthly average wind speed readings from three distant

meteorological station (20, 30 and 60km) using wind index method and this extended

dataset was used as long term reference. Also, it would have been possible to extend this

dataset with some other online data and make use of this dataset as a long term

reference. However this correlation has again to be done with the online sources and

making several long term correlations to obtain one long term reference data increases

the uncertainty.

3.4.1 Data Validation

For simulations, only the measurements from the top anemometers are used. The

measurements from different heights are used to cross-check and detect abnormal wind

speed readings. For M3, averaged value of two top anemometer’s wind readings is used.

Before using the time series, they are inspected to detect faulty data and decrease the

uncertainty that might possibly be caused from the data quality. Datasets are investigated

using WindPro and Windographer. WindPro is used to detect erroneous data, duplicates

and data lies out of the normal ranges. Range test is applied to detect wind speed data

more than 30m/s. In all time series no wind speed more than 30m/s is encountered. Later

on, the erroneous data giving pre-set error values are removed. M1 and M2 datasets

contain some erroneous data giving the error code (-99.99). M1 and M2 contain 88 and

272 invalid data points (the error value) respectively. These data are discarded from the

time series. The used period of the time series and the data coverage (percentile coverage

of a full year) after the removal of faulty data can be seen in Table 3.

36

Table 3: The used period of time series and the coverage after the removal of invalid

data

Dataset From To Coverage

M1 12/02/07 13:40 12/02/08 13:30 100%

M2 14/02/07 11:10 14/02/08 11:00 99.5%

M3 09/06/09 00:10 09/06/10 00:00 100%

Note that, M1 and M2 measurements are carried out for 13 months and the invalid data

in the time series for M1 were present at the beginning and the end of dataset therefore

after the removal of invalid data there was still one full-year data available. Thus the

coverage of M1 dataset is not changed due to the erroneous data while the coverage of

M2 is slightly reduced.

Windographer is used to inspect the shading effect for M3 which has two anemometers

at same height. Windographer allows its users to auto flag the tower shading if two

anemometers at same height are present. It detects the wind directions where the tower

shading occurs (see Appendix A). Two anemometers at 60.1m have really close wind

speed readings, the difference between two anemometer’s average wind speeds for the

entire period is only 0.05 m/s. Tower shading flag is applied to the two anemometers, it

slightly increased the average wind speed from anemometer_1 by around 0.009% while

decreasing the average wind speed from anemometer_2 by 0.009%. Brower (2012)

defines averaging the wind speed readings as an effective method to mitigate tower

shading effect. Considering this, two time series from the two anemometers are averaged

and the resulting average wind speed was 0.01m/s more than that of anemometer_1. It is

concluded that, for M3, tower shading has no remarkable impact on the results since it

has almost not changed the average wind speed.

Icing effect is investigated by checking the dataset manually. The average temperature

values of M1, M2, and M3 time series are found to be 8.4, 8.6 and 7.6 °C respectively.

37

Data points that the temperature is below zero are checked. Consistent low wind speed

readings during the low temperature values are not encountered therefore a relation

between the temperature and wind speed readings could not be detected. Thus, no

correction is made for icing. Note that, a loss of only 0.5% was estimated for the losses

caused by ice and dirt accumulation on the blades. This value was determined depending

on the observations of the local climate.

3.5 Production Data

Production data is needed to make comparisons with the real case. Production data is

available for three years from January 1st 2013 to December 31

st 2015 and shows big

monthly and annual variations. Monthly fluctuations of the operational data (real data)

for the available three years can be seen in Figure 9.

Figure 9: Monthly fluctuations of power production for three years

These values are actual production numbers of the farm including the all losses.

Production data varies a lot between the same months of different years. For example,

0.0

1,000.0

2,000.0

3,000.0

4,000.0

5,000.0

6,000.0

7,000.0

8,000.0

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

MW

h

2013

2014

2015

38

production data of Dec. 2013 is 33% more than Dec.2014. In June 2014, capacity factor

goes as low as 9.7% while it is 21.1% in June 2015. As can be seen from the graph,

monthly and annual fluctuations are significant for this case and since the wind

measurement might be carried out in any year, the long term correlation gains

importance for the accurate AEP estimations. The discrepancy between the original AEP

estimation and the actual production values can be seen in Table 4.

Table 4: Discrepancy between the original AEP estimation and the operational data

2013 2014 2015

Operational Production Data (GWh) 45.98 41.57 46.92

Actual Capacity Factor 26.23% 23.82% 26.73%

AEP Estimated (GWh) 53.4 53.4 53.4

Capacity Factor Estimated 30.5% 30.5% 30.5%

Discrepancy 16.1% 28.5% 13.8%

Average Discrepancy 19.1%

3.6 Measure Correlate Predict (MCP)

Due to the absence of sufficiently long measurements, online datasets available in

WindPro/EMD database are checked and several reanalysis datasets in the vicinity of

wind farm are found. Reanalysis is a scientific method for developing a comprehensive

record of weather and climate fluctuations over time and it is being used increasingly in

commercial and business applications in sectors such as energy, agriculture, water

resources, and insurance (CIRES n.d.). Especially, in the absence of the sufficiently long

term measurements, these datasets are widely used as long term reference datasets.

Reanalysis relies on the constant data assimilation to create worldwide observational

data and the sources of this observational data is rather wide such as surface weather

stations, weather balloons and, airport and satellite measurements (Liléo et al. 2013).

Available datasets are sorted out considering the coverage and distance from the site.

Liléo et al. (2013) suggest the choice of 15 to 20 years long term reference data therefore

datasets which have records less than 20-year period are eliminated. According to

Brower (2012), in complex terrain or near coastlines, the ability to reliably extrapolate

39

the information beyond a station’s immediate vicinity is more limited. Thus, long term

dataset search is limited to the range of 15km from the site. Two adequate datasets were

found. First one, MERRA dataset originates from the Global Modelling and

Assimilation Office of NASA. The MERRA analysis are being conducted with the

GEOS-5 Atmospheric Data Assimilation System (ADAS) and the model grid is 0.5

degree latitude and 2/3 degree longitude (NASA 2012). The location of the data is five

km away from the site and measurement height is 50m. The data is available from 1979

to present. Correlation rate of on-site measured wind data and MERRA dataset are found

to be quite decent which is also interpreted as a sign of healthy long term correlation.

The other dataset, CFSR-E is developed by NCAR. The data is available for the period

from 1979 to present. The spatial resolution is 0.5 degrees. The location of data is

approximately 11km from the site and the measurement height is 10m. Details

regarding to the mentioned data sets can be seen in Table 5.

Table 5: Details of long term reference datasets

Data Name Approx. Distance (km) Used Period Data interval Data height (m)

MERRA 5 1996-2016 1 hour 50

CFSR-E 11 1996-2016 1 hour 10

Long term correlation is carried out using the WindPro MCP module for the 60m key

height which is the hub-height of the investigated turbines. Thøgersen et al. 2007

suggests the use of matrix and regression model for the sites where both local site data

and long term reference data are available as detailed time series which is the case for

this study. The matrix method is used for correlation. The weighted mean of a sector

wise correlation is a value showing the quality of correlation which is obtained by

weighting the concurrency of wind data of all sectors. This rate will be further

mentioned as the correlation coefficient and this rate for M1, M2 and M3 with MERRA

dataset is calculated as 0.88, 0.86, and 0.84 respectively. The correlation coefficient

calculated based on hourly values is first and foremost a measure of how well the short-

term fluctuations in the reference and in the measured wind speed data agree in phase

40

(Liléo et al. 2013). These numbers indicate a good correlation between on-site

measurement and long term reference data. However, a good correlation coefficient

does not always mean that the representativeness of the long term reference data is good.

A long term data might follow different patterns than the on-site measurements and

might still be more representative of the climatology. However, the correlation rate is the

most common measure that we can assess the performance of a correlation and a decent

coefficient is usually a sign of a good correlation.

Correlation coefficients of the CFSR-E dataset with M1, M2, and M3 are 0.85, 0.84, and

0.81 respectively which is slightly lower than that of the MERRA. Detailed results

regarding the correlation coefficients can be seen in Appendix B. GH and Partners

(2009) suggests having at least a correlation coefficient of 0.6 to use a dataset as long

term reference. It can be seen that both long term references have quite decent

correlation coefficients. In order to assess the influence of long term references on the

AEP estimation, both dataset will be separately used as long term reference and the

results will be discussed. According to NCAR (2016), relatively few evaluations of

CFSR dataset have been conducted therefore the performance is not well-known.

According to Liléo et al. (2013), MERRA dataset is one of the most proper global

dataset for the use as long term reference data for the low-complexity sites among other

analysis datasets. Moreover, the location of MERRA data is closer to the site and the

measurement height is closer to the hub-heights of the turbines. Due to the

abovementioned reasons, MERRA data is chosen as the primary long term reference for

the final results. All simulations will be carried out using both dataset as long term

reference but the assessed uncertainties will only be applied to the AEP results obtained

using MERRA dataset.

3.7 Wind Farm Modelling Tools

WindPro and WindSim are used to model the site and simulate the micro wind

climatology. For AEP calculations, the energy module in WindSim and the park module

41

in WindPro are utilized. The park module employs the WAsP in WindPro interface for

AEP calculations. As mentioned before, WAsP uses a linear flow model which does not

account for the flow separation that high complexity sites and varying roughness

situations induce (Wallbank 2008). Therefore WAsP is expected to estimate a higher

gross energy production compared to WindSim.

3.7.1 WindPro

As it is mentioned in the previous chapters in details, WindPro is used to handle the

digitized topography data, data validation and long term correlation. These processes are

explained in separate chapters because the resulting data are used in the both simulation

tools (WindSim and WindPro) and these processes can constitute a reason for the

discrepancy by themselves.

After creating 6 climatology datasets by long term correlating M1, M2 and M3 time

series with MERRA and CFSR-E reference data using Matrix method in MCP module,

The STATGEN (Statistic generation) module is used to create wind statistics from the

MCP calculations. Therefore 6 wind statistics were created. Wind statistics are needed

for WAsP calculation because the MCP module performs the long term correlation

without taking the topography into account. The STATGEN module generates wind

statistic taking account the elevation, roughness, the obstacles and the climatology data.

After the wind statistics are created, we can proceed to the next step which is the

calculation of AEP using the PARK module. WAsP 11 is employed in the WindPro

PARK energy calculation module for the AEP estimations. In this module, the wake

decay constant, wake model, desired wind statistics, power curve and air density

correction methods are chosen. Simulations are repeated with the five available wake

models. The wake decay factor is chosen as 0.075. This factor is the default value of the

WindPro for on-shore applications and according to Renkema (2007), this value is an

adequate choice for the land cases.

42

Air density correction is made benefiting the data from a nearby meteorological station

which is approximately 30km from the site thus air density at each turbine’s hub height

is calculated. Air density correction caused 1.4% reduction in AEP. Wake losses are

depending on the used wake model and will be investigated in the results chapter. The

additional losses are also estimated in the following chapters and deducted from the

production number along with the calculated wake losses and thereafter the Net AEP

values are obtained.

3.7.2 WindSim

WindSim is used to simulate the case with different wake models and long term

references. Windsim consists of 6 modules. The relevant modules and their functions are

explained along with the followed steps.

3.7.2.1 Terrain Module

A detailed terrain model is needed to model the flow over the terrain. Moreover on this

terrain, WindSim requires defining a horizontal and vertical grid to solve the 3D RANS

equations for the nodes of the grid. Parameters of horizontal and vertical grid, grid

resolution and refinement are arranged in this module. Also, additional features like

terrain smoothing are available in this module. The terrain smoothing model is usually

used if the model is not able converge due to the abrupt changes in the terrain which is

not the case for this study.

20x20km elevation grid data and roughness lines are merged using WindPro and

exported as a combined map file and then this file converted into a gws file in order to

input it in WindSim. For the 3D grid, a maximum number of 1.000.000 cells are chosen

and 25 cells in the vertical direction are assigned. The number of the cells in the x, y and

z direction and grid spacing can be seen in Table 6.

43

Table 6: Grid spacing and the number of cells

Easting Northing Z Total

Grid spacing (m) 23.4-457.7 23.4-454.3 Variable -

Number of cells 200 200 25 1000000

Refinement is made to achieve a denser distribution of the cells in a specified area. In

this work, the core of the site (3x3km) where the wind farm takes place is refined.

Schematic view of horizontal and vertical grid can be seen in Figure 10.

Figure 10: Horizontal grid resolution (left) and schematic view of the vertical grid

resolution (right) Source: WindSim

The vertical grid extends 1679m above the terrain and it is refined towards to the

ground. There are 6 cells in the z direction below the hub-height of the turbines. This

refinement is done to achieve higher accuracy. The maximum and minimum height

distribution of the first ten nodes in the vertical grid relative to the ground level can be

seen in Table 7.

Table 7: Distribution of first 10 nodes in z direction

1 2 3 4 5 6 7 8 9 10

z-dist. max (m) 3.3 10.0 16.7 23.6 34.0 51.0 74.4 104.4 140.8 183.7

z-dist. min (m) 3.3 10.0 16.7 24.1 36.1 55.6 82.5 116.9 158.7 208.0

44

3.7.2.2 Wind Fields

In the wind fields module, RANS equations are numerically solved for every node

therefore increasing cell numbers increases the needed time for the simulation. The flow

on the terrain is simulated in this module according to the set parameters such as

boundary layer, speed assumption above the boundary layer, boundary conditions,

turbulence closure and the solution method. The boundary layer height is the highest

altitude that the flow is assumed to be influenced by the surface. For all of the

simulations run in WindSim, the height of boundary layer is set as the default value

500m. The wind speed above the boundary layer which stands for the geostrophic wind

speed is set as 10m/s which is also a default value. The standard k-epsilon model is

applied as turbulence closure. Two types of boundary conditions at the top are available

in WindSim namely, fixed pressure and non-frictional wall. The fixed pressure approach

is suggested for complex terrains by Windsim thus this approach is specified for the

boundary conditions. The RANS equations are solved with the GCV (General

Collocated Velocity) method which was also used in the studies of Gallagher (2014),

Nilsson (2010) and Wallbank (2008). In these computations, rotors of wind turbines are

treated as actuator disks. Twelve simulations were performed for twelve sectors. For

every sector, a number of iterations are performed until the results are converged. The

iteration numbers for every sector to reach to the converged value can be seen in

Appendix C.

3.7.2.3 Objects

In this module, wind turbines and the climatology objects are located on the terrain, see

Figure 11. Also, details regarding to the turbines such as hub height, rotor diameter and

power curve are entered in this section. Wind turbine details are imported from the

WindPro database. After the objects module is run it enables the following modules to

be run. Also, in order to calculate the wake effects caused by the presence of the

adjacent farm, turbines of the neighboring farm are located on the terrain.

45

Figure 11: Turbines and met masts located in the objects module Source: WindSim

Note that, met masts located on the terrain do not represent the on-site measurements.

They contain the climatology data which is created by correlating the on-site

measurements with the local wind climate (long term dataset) in WindPro. The

climatology data is converted from the txt. file to a tws. file using Windographer for

WindSim. The climatology datasets contain 20 years of data from 01/1996 to 01/2016.

Table 8 and Table 9 show the details regarding to climatology data as they are perceived

by WindSim.


MERRA dataset Source: WindSim Climatology name Representative period Measurement height (m) Average wind speed(m/s) M1/MERRA 01/01/1996 - 01/01/2016 60.10 9.11 M2/MERRA 01/01/1996 - 01/01/2016 60.10 7.67 M3/MERRA 01/01/1996 - 01/01/2016 60.10 7.25


CFSR-E dataset Source: WindSim Climatology name Representative period Measurement height (m) Average wind speed(m/s) M1/CFSR-E 01/01/1996 - 01/01/2016 60.10 8.96 M2/CFSR-E 01/01/1996 - 01/01/2016 60.10 7.51 M3/CFSR-E 01/01/1996 - 01/01/2016 60.10 7.20

46

3.7.2.4 Wind Resources

Wind resources module is run to visualize the average wind speed distribution over the

terrain. The flow influenced by the terrain was simulated in the wind fields module.

After inserting the climatology data in the objects module, now the numerical mean

wind speed distribution over the terrain can be simulated. As mentioned above, three

climatology datasets are available and the result is based on all the climatology data,

obtained by the inverse distance interpolation of the climatology objects (Meissner

2010). The resulting wind resource map of the site can be seen in Figure 12.

Figure 12: Mean wind speed at 60m based on 3 climatology objects (grey points

represent the climatology object locations) Source: WindSim

47

3.7.2.5 Energy

The energy module can be run after the objects module to calculate the AEP. This

module is run with three available wake models (Jensen, Larsen and Ishihara) and air

density correction function. The air density is corrected individually according to the

altitude of hub-heights (individual correction 1). The density at hub-heights is calculated

considering the elevation change above the sea level. According to WindSim, this

correction provides a good long-term average value of air density in moderately complex

areas. Note that, different air density correction methods are used in WinPro and

WindSim. WindSim flow model and the climatology data are used to simulate the free

stream (not wake affected) wind speed distribution. Numerical flow modelling is

employed to calculate wind speed and direction at the hub heights originating from

measurement points. The gross AEP is calculated using this wind speed distribution at

the hub height of each turbine location and the provided power curve. Results are

acquired using the frequency distribution which is obtained from the frequency tables of

the climatology data.

3.7.2.6 Grid Independency Study

The grid resolution impacts the results acquired using WindSim. A grip independency

study is carried out in order to see whether the results are grid independent or not.

Simulations started with 500.000 cells and are carried out until the max allowable cell

4.000.000 is reached. No parameters are changed except the cell numbers. The gross

energy production changes in results obtained with different cell number are in the

vicinity of 0.1%. Therefore it can be said that used cell number has no considerable

impact on the results. Deviations in the calculation of Gross AEP (MWh/y)

corresponding to the different cell numbers are presented in Figure 13.

48

Figure 13: Changes of AEP for different cell numbers in WindSim for all turbines

3.8 Losses

As it is mentioned in Chapter 2, the wake losses are calculated by employing the wake

models available in WindPro and WindSim. While the rest of the losses that wind farm

experiences are estimated based on the provided numbers by the original WRA

document and the literature review. In this work, availability, electrical losses,

environmental losses and the turbine performance losses are evaluated. An estimation

regarding the curtailment losses is not performed due to the absence of relevant data.

Curtailment losses are very case dependant and vary greatly for different wind farms

therefore influence of this loss on the results is suggested as a further research.

Availability comprises the turbine availability, grid availability and the substation

maintenance losses. Depending on the original WRA document, 5% of total availability

losses are assumed. Considering the literature review, total electrical losses comprising

the transmission losses and the wind farm consumption loss are assumed to be 1% based

on the literature review. Environmental losses due to the ice and dirt accumulation were

estimated considering the local climate and conditions of the site by the company and it

was estimated to be 0.5%. This rate is assumed to be true for this paper as well. For the

49

turbine performance losses, in the original WRA, the loss due to high wind speed

hysteresis is estimated as 0.6% however the power curve adjustment losses is not

considered. Mortensen (2011) suggests 1 to 2% of total turbine performance losses to

consider for the modern wind farms. In this paper total performance losses is assumed to

be 1.5%. A summary of assumed losses can be seen in Table 10.

Table 10: Assumptions of additional losses

Loss category Loss Types Estimation

(%)

Availability

Turbine availability

5% Grid availability

Substation maintenance

Electrical

Efficiency Transmission efficiency

1% Wind farm consumption

Turbine

Performance

Power curve adjustment 1.5%

High-wind hysteresis

Environmental Blade degradation and fouling

0.5% Icing or temp. Shutdown

Curtailment

Losses

Wind sector management - Noise, visual and

environmental

Total Losses 8%

Note that, some of the losses such as power curve adjustment were not considered in the

original WRA document and it was suggested to consult the manufacturer to determine

this loss factor. However it is known that, losses due to the power curve adjustment is

expected unless the wind turbine operates under the standardized test conditions. For this

paper, not-assessed losses are quantified depending on the literature review. Therefore,

the total additional losses are estimated as 8% in this paper while the considered losses

in the original WRA were 6.6% in total. This estimation of the additional losses will be

applied to the Gross AEP along with the calculated wake losses to obtain Net AEP

values.

50

CHAPTER 4. APPLICATION OF THE METHODOLOGY AND RESULTS

In this chapter, the results of the simulations are presented to the reader. Simulations are

performed with the climatology data obtained through the correlation of three available

time series (M1, M2 and M3) with two different long term reference datasets (MERRA

and CFSR-E). This is made to assess the influence of different long term references on

the results. Simulations are also executed with the all available wake models in

WindSim and WindPro. This is done in order to assess the performance of the wake

models and the compatibility of the results with the literature review. Since there are

variables to consider, results are analyzed under several subchapters.

4.1 Interpretation of wake losses

Results obtained using different wake models and the distribution of the wake losses per

sector are evaluated in this section.

4.1.1 Influence of wake models

Wake losses are found to be significant within the wind farm. These numbers are the

consequences of the spatial array of the Wind Farm A. The results obtained with

employing different wake models are consistent with the studied literature review for the

analysis of the wake models. The wake losses are calculated employing the wake

models available in WindPro and WindSim using only the default parameters. The

results can be seen in Table 11.

Table 11: Calculated wake losses

Simulation Tool Wake Model Wake losses (%)

(M1,M2 and M3/ MERRA)

Wake losses (%)

(M1,M2 and M3/ CFSR-E)

WindSim 7.0. Jensen 8.4 8.6

WindSim 7.0. Larsen 4.5 4.6

WindSim 7.0. Ishihara 8.4 8.6

WAsP 11 Jensen 8.2 8.4

WAsP 11 Jensen 2005 7.2 7.3

WAsP 11 Larsen 5.4 5.5

WAsP 11 Larsen 2008 5.4 5.5

WAsP 11 Eddy Viscosity 5.0 5.1

51

The multiple wake losses (overlapping wakes from different turbines) are calculated

using the wake combination model of ‘sum of squares of velocity deficits’ in WAsP and

WindSim. As can be seen, the wake losses are calculated for the two different

climatology file. The only difference between the two different climatology data is the

long term reference used (MERRA and CFSR-E). Different long term references change

the calculated average wind speed at hub-height. Since all the wake models take the

incoming wind speed into account, slight deviations occur between the losses calculated

with different climatology data.

Also, the software used has an influence on the wake losses. The Jensen wake model

gives quite similar results for WindSim and WindPro while the losses calculated with the

Larsen model differ from each other. This can be explained with the different flow

model approach which leads to different calculated turbulence intensity values. The

Jensen model does not directly take the turbulence intensity into account while the

Larsen model is directly influenced by the turbulence intensity. Orographic turbulence

induced due to the complexity of the terrain might be accelerating the wake recovery for

WindSim.

Even though the different climatology data and the use of different software caused

slight deviations in the wake losses, it can be easily seen that the main determinant factor

for the losses is the wake model. It is needed to agree on a wake model for further

simulations. The results of the studied researches are taken as a reference while choosing

an applicable wake model. Renkema (2007) claims that the Larsen model performs

badly in WindPro and underestimates the wake losses due to the shallow and wide wake

estimations. According to Sørensen et al. (2008), if the default parameters are used, the

Eddy-Viscosity model is inclined to lead the lowest wake losses in WindPro. Taking a

glance to the results of this study, it can be seen that the Larsen wake model calculates

the lowest wake losses just after the Eddy-Viscosity model in WindPro. Also, as

52

expected, the original Larsen and the modified version (Larsen 2008) display exactly the

same results. Moreover, the Larsen model calculates even less wake losses in WindSim.

Also, as it is stated by Renkema (2007), when the wakes from different turbines overlap,

the modified Jensen wake model (Jensen 2005) shows rather stable losses while the

original Jensen wake model leads to increased losses. In this work, Jensen 2005 results

in approximately 1% less wake losses compared to the original Jensen wake model for

all the simulations performed. In WindSim, the Jensen and Ishihara wake models end up

with a really close loss percentile.

In the literature review, the Jensen wake model was claimed to provide decent results

with a fine wind decay factor and it is the suggested model by WindPro. Due to its

availability in both software and considering the literature review, only the original

Jensen wake model with the default suggested wake decay constant for the land cases

(0.075) is used for the following simulations. This decision is taken in order to narrow

the area of investigation and prevent confusion which might occur due to the existence

of too many variables.

4.1.2 Sector-wise wake losses

According to the production analysis of WindPro, the major wind direction is south-

south west and the wind farm is impacted to the highest wake losses from the south-

south east direction. Annual wake losses for the SSE direction are calculated to be

around 870MWh which amounts to 15% of total wake losses from 12 sectors. This is

probably due to the adjacent wind farm, located to the south and south east of the farm.

Annual energy yield and wake losses per sector can be seen in Figure 14.

53

Figure 14: Energy yield and wake losses per sector Source: WindPro

4.2 Simulation Results

As it is mentioned in the previous chapters, six climatology data are created by

correlating three on-site measurements (M1,M2 and M3) with two long term datasets

(MERRA and CFSR-E). Some abbreviations are assigned to the climatology data to

make the follow-up of the results easier. For instance, M1/MERRA defines the

climatology data obtained using the M1 time series as an on-site measurement and the

MERRA data as long term reference.

First, six climatology data obtained are used in the simulations individually. Simulations

are performed to identify the influence of WAsP and WindSim and two different long

term reference data (MERRA and CFSR-E) on the AEP. The long term reference data is

54

found to be one of the most effective parameter on the AEP estimations. Results of the

simulations can be seen in Table 12. The Gross AEP represents the AEP estimation

without any losses and the Net AEP is obtained after the wake and additional losses are

deducted from Gross AEP. Details regarding the climatology data can be seen in

Appendix D.

Table 12: Simulation results for all climatology data in WindSim and WindPro

Climatology name

WindSim 7 Gross AEP (GWh/y)

WAsP 11 Gross AEP (GWh/y)

WindSim 7 Net AEP (GWh/y)

WAsP 11 Net AEP (GWh/y)

M1/MERRA 65.86 68.54 55.8 58.3

M1/CFSR-E 63.50 65.97 53.7 55.9

M2/MERRA 59.41 59.70 50.0 50.3

M2/CFSR-E 56.17 56.90 47.1 47.9

M3/MERRA 55.27 55.22 46.4 46.3

M3/CFSR-E 52.97 53.87 44.4 45.2

Wind farm A consists of 8 wind turbines but their distribution on the site is not uniform,

turbines are located as a cluster of 6 and 2 turbines with an approximate 1-1.5 km

distance to each other. The measurement locations of M2 and M3 are near the first

cluster (6 turbines) with a 400m distance to each other. M1 takes place near the second

cluster (2 turbines). The distance from M1 to M2 and M3 is approximately 1.5-2km. All

of the time series follow compatible patterns with the long term references. Therefore

the positioning explains the difference of the mean wind speed between the M1 and M2

measurements. Note that, the M3 time series contain data from 06/2009 to 06/2010 while

the M2 and M1 contain wind data from 02/2007 to 02/2008.

WindSim and WindPro are able to handle several climatology files for the AEP

calculations. This process is done by inverse distance extrapolation of the climatology

objects considering the measurement points and the turbine locations. Since the turbines

and the measurement locations are scattered on the site and have different proximities to

55

each other, performing simulations based on the all climatology data is expected to lead

to more accurate results. Also, according to WindSim, increasing the amount of wind

data improves the accuracy of the numerical results. The first set of simulations is

performed using the M1/MERRA, M2/MERRA and M3/MERRA climatology data and

the second one using the M1/CFSR-E, M2/CFSR-E and M3/CFSR-E in WindPro and

WindSim. As it is decided in the previous section, only the original Jensen wake model

is employed for the wake losses and 8% of additional losses are applied. Therefore the

only variables differing in the simulations are the long term reference data and the

software. Results can be seen in Table 13. The average wind speed stands for the

average of the not-disturbed mean wind speeds at the hub height positions of all turbines.

Table 13: AEP results with MERRA and CFSR-E datasets as long term reference

Simulation Tool

Gross AEP (GWh/y)

Average wind speed (m/s)

Wake losses

(%)

AEP with wake losses

(GWh/y)

Capacity factor(%)

Net AEP (GWh/y)

Climatology M1,M2 and M3/ MERRA as long term reference

WindSim7.0. 60.3 7.9 8.4 55.3 29 50.9

WAsP 11 61.5 7.9 8.2 56.4 29.6 51.9

Climatology M1,M2 and M3/CFSR-E as long term reference

WindSim7.0. 57.7 7.7 8.6 52.7 27.6 48.5

WAsP 11 58.9 7.7 8.4 54 28.3 49.7

As can be seen, the two different long term reference datasets have a significant

influence on the results. Changing the long term reference changes the results by 4.5-5%

for both of the software while the deviation occurred due to the use of different software

is in between 2-2.5%. This reminds us of the importance of a decent long term

correlation especially for the sites where the annual wind deviation is high. The

production data of 2013, 2014 and 2015 are available. Discrepancy of the simulation

results based on the average production of three available years (44.8GWh) can be seen

in Table 14.

56

Table 14: Discrepancy of the results compared to the operational data

Note that in the original WRA, WAsP was used for the simulations and the long term

reference data was created extending the M4 dataset by correlating it with the monthly

average wind speed readings from three distant meteorological station using wind index

method. The original WRA estimated the gross AEP as 61.5GWh/y and the net AEP as

53.4 GWh/y. If the MERRA dataset is taken as long term reference, WAsP makes a

close AEP estimation to the result of original WRA. While using CFSR-E as long term

reference decreases the discrepancy to 10.9%. Employing WindSim leads to the closer

estimations to the operational data for all of the cases by approximately 2-2.5% on

average. On the other hand, if CFSR-E dataset is used as long term reference and

WindSim as the simulation tool, it can be seen that the discrepancy rate reduces as low

as 8.2%. It is not easy to judge which long term reference represents the climatology

better since the correlation coefficients from both dataset are quite decent. However it

now can be said that, the discrepancy cannot only be explained by the use of WAsP out

of its suggested envelope since the influence of the software is found to be much less

than the overall discrepancy.

4.3 Uncertainty Assessment

Uncertainties were studied in chapter 2. Now the assessed uncertainties will be applied

to the results. As it can be remembered, the MERRA dataset was chosen as the primary

long term reference mainly due to its proximity to the site and the measurement height.

Also, WindSim is found to provide more accurate results due to the complexity of the

site. In order to decrease the modelling uncertainty, the uncertainties assessed will be

Net AEP (GWh/y) Average discrepancy

WindSim 7.0.0. (MERRA) 50.9 13.6%

WAsP 11. (MERRA) 51.9 15.8%

WindSim 7.0.0. (CFSR-E) 48.5 8.2%

WAsP 11. (CFSR-E) 49.7 10.9%

57

applied to the result obtained using WindSim. WindSim estimated the Net AEP as

50.9GW/y using the climatology data M1/MERRA, M2/MERRA and M3/MERRA.

Uncertainties are identified and evaluated depending on the provided information by

manufacturers, simulation results and the literature survey.

The accuracy of used anemometers (Vector A100L2 Anemometer) is defined as %1

(±0.1m/s) of the speed readings by the manufacturer (Campbell Scientific 2016).

Uncertainties in sensor calibration and the boom and mounting effects are decided as

0.5% each depending on the literature review. Thus the total measurement uncertainty is

assigned as 2%.

According to GH and Partners (2009) , if the correlation rate is in between 0.8 and 0.9,

correlation uncertainty can be taken as 1-2%. There is also a prediction uncertainty

(using past as a predictor of future). The total uncertainty in the long term correlation is

taken as 3%, also this estimation corresponds to the suggestions of Liléo et al. (2013).

The typical range for modelling uncertainty was suggested as 3-6% by Lira et al. (2014).

It is assumed that the use of WindSim decreases the uncertainty in modelling. Also

availability of the measurements at hub height decreases the uncertainty for vertical

extrapolation which is the case for this study. The lower bound 3% is assumed to be the

modelling uncertainty. Uncertainty of wake losses is not considered and also future

climate change uncertainty is not taken into account. The power curve uncertainty is

assumed to be 5% depending on the literature survey. Also, while determining the

additional losses, literature survey is benefited along with the specifically assessed losses

for the wind farm in case. Therefore 5% of uncertainty for the additional losses is

assumed.

58

Table 15: Assumed uncertainties associated with the WRA and AEP calculations

Win

d R

eso

urc

e U

nce

rtai

nty

Measurement Uncertainty

Sensor Accuracy 1.0 %WS

Sensor Calibration 0.5 %WS

Boom and Mounting 0.5 %WS

Long Term Correction Correlation Uncertainty

3.0 %WS Past-Future Relation Uncertainty

Modelling Uncertainty Wind Flow Model 3.0 %WS

Vertical & Horizontal Extrapolation

Ener

gy A

sses

smen

t U

nce

rtai

nty

Power Curve Uncertainty 5.0 %AEP

Metering Uncertainty - %AEP

Uncertainty of Losses

Availability Losses

5.0 %AEP Electrical Losses

Environmental Losses

Other losses

Year to year variability of wind speed is in between 4% and 8% according to the used

long term reference dataset therefore 6% standard deviation of annual wind speed is

estimated. Also, in the original WRA document, annual wind variation is assessed as

6%. These uncertainties are applied to the results using WindPro Loss & Uncertainty

module. In this module, the assessed loss percentile and loss type are specified. These

uncertainty percentages are the limits of the errors based on the experiments and

literature review. Depending on the defined limits, production probabilities are

calculated. In the table below, P50, P75 and P90 values stand for the production values

which are likely to be exceeded with %50, %75 and %90 probabilities respectively.

Probability of exceedance for 1, 5, 10 and 20 years average are calculated and can be

seen below.

Table 16: Probability of exceedence for different time intervals

Probability of exceedance [%]

1y [GWh/y] 5y [GWh/y] 10y [GWh/y] 20y [GWh/y]

50 50.9 50.9 50.9 50.9

75 45.6 47.0 47.2 47.3

90 40.9 43.4 43.8 44.0

59

Operational production data is available for 2013, 2014 and 2015. In 2014, the

operational production rate of the farm was 41.6 GWh/y and this was the lowest

production value among the available 3 years. The average production value of the

available 3 years was 44.8GWh/y. As we can see from the Table 16, P90 value of one

year production is 40.9GWh and this number is 43.4GWh/y for the 5 years average.

Even though the production values are expected to be more than these numbers with a

90% probability, the actual production data seem to be still within the bound of

possibility according to the new AEP estimation.

CHAPTER 5. DISCUSSION AND ANALYSIS

Simulations performed using WAsP in WindPro interface and WindSim revealed that

the bias caused from the use of different software accounts for 2-2.5% and might only

partially explain the discrepancy since the discrepancy investigated in this work is much

larger than this number. This deviation might be caused from the different wind flow

modelling approach and the different calculated wake losses since the additional losses

are applied to the results from both software as a fixed number. However, the calculated

wake losses using the Jensen wake model are quite similar in the both software.

Generally, WindSim calculated only 0.2% more wake losses compared to WAsP.

Therefore it can be suggested that this bias mainly occurs due to the mid-complex

property of the site which WAsP cannot effectively handle.

On the other hand, wake losses obtained with different wake models show large

variations therefore the use of an unrepresentative wake model might lead to significant

errors in the AEP estimation. Especially the Larsen model estimates approximately 4%

lower wake losses compared to the Jensen and Ishihara model in WindSim. These big

differences are probably encountered because of the close spacing of the wind turbines

(between 2.8 and 4.1 D). The Jensen and Ishihara model in WindSim calculates

consistently the almost same wake losses for the investigated farm. Also, the wind farm

60

experiences the highest wake losses from the south and south east direction. An adjacent

wind farm is located at the south and south-east part of the investigated wind farm which

might explain this situation.

According to Liléo et al. (2013), 3 to 7% variation of annual mean wind speeds

corresponds to approximately 8 to 18% variation of AEP. For the case of this study,

annual wind variations are found to be high and it can also be seen from the fluctuations

of the AEP values. The production rate of 2014 is 12.9% less than that of 2015. These

high fluctuations in wind, makes long term correlation a crucial point for accurate long

term estimations. However, the choice of a different long term reference had a

significant impact on the results. Two different long term reference data (MERRA and

CFSR-E) led to approximately 4.5-5% difference in the results and this discrepancy

occurred even though both of the datasets cover the same period and have quite similar

and decent correlation coefficients with on-site measurements. According to Liléo et al.

(2013), correlation coefficients do not provide an ideal measure for the

representativeness of long term reference. So here we can conclude once more that, a

decent correlation coefficient does not always mean that the climatology of the site is

reflected into calculations very well. Considering the big annual fluctuations of the

annual wind speed and the impact of long term correlation on the results, it can be said

that long term reference and correlation might be a probable reason accounting for a

considerable part of the discrepancy.

Furthermore, when the only long term correlated M3 time series is used in the

simulations, results illustrates really low discrepancy rates (see Table 12). WindSim

simulations using the M3/MERRA and M3/CFSR-E climatology datasets estimate the

AEP as 46.3 and 44.4 GWh/y respectively. The first case overestimates the average

AEP of the available 3 years by 3.3% while the second one underestimates it by

0.9%.When the long term datasets are examined, it can be seen that the period of the M3

measurement (06/2009-06/2010) witnessed lower average wind speed than that of M1

61

and M2 (02/2007-02/2008). Wind conditions in the period that M3 measurement was

carried out might be more similar to the years when production numbers were measured

and the long term correlation might not be effectively mitigating the impact of this low–

high wind speed years.

Finally, note that these comparisons are made with the available three years of

production data however AEP estimation is the average value of 20 years estimated

production. Therefore the comparability of these two results is limited. So there is a

possibility that, the available 3 years might be low wind speed years and the average

production value might increase for a 20 years period. When the uncertainties are

applied to the results, P90 value for a single year and five years average are calculated as

40.9 GWh/y and 43.4GWh/y respectively. On this basis, it can be said that the

production data of three years are still within the bound of probability for the new AEP

estimation obtained using WindSim with MERRA dataset as long term reference.

Limitations

The thesis is considerably limited by the time constraint. Also, being far from the

investigated site so not being able to make on-site investigations in order to

evaluate the current state of the wind site is a hindrance for the thesis. Online

sources are benefited to identify the sheltering objects. However this evaluation

of the site relies on the up-to-dateness of the online sources. Changes in the

roughness profile of the site which are reflected to the online databases yet might

cause a discrepancy between the assessment made in this paper and the reality.

Curtailment, availability and outages are included in the production data and

their variations among the months and years are unknown. Availability losses are

estimated depending on the original WRA document and literature review and

this number comprises outages as well. However, curtailment losses are not

assessed either in this work or previously made WRA. Since the distances among

62

the turbines is sometimes as low as 2.8D, curtailment losses due to the wind

sector management might be contributing to the underperformance and due to

absence of related data it could not be investigated.

The production data was available for the whole wind farm but the data for

individual wind turbines was not present. The data from individual turbines

might allow the author to make a more precise work for the wake losses.

Production data is available only for 3 years and reflects a limited period

therefore the comparability of the AEP estimations with the actual production

data is limited since the AEP estimations are the estimated average production of

20 years.

CHAPTER 6. CONCLUSIONS

To identify and quantify the factors which might account for the discrepancy observed

between the pre-construction AEP estimation and the operational production data the

performance of an actual wind farm was investigated using three years of operational

data. The average AEP of these three years indicates a 19.1% underperformance

compared to the AEP estimation obtained from a wind resource assessment (WRA)

performed prior to the construction.

It is noticed that, the original WRA document did not include loss factors such as power

curve adjustment and curtailment losses. In the present study, power curve adjustment

losses are estimated depending on the typical ranges of losses for modern turbines while

curtailment losses are found to be case specific and not assessed in this paper as well.

The additional losses are estimated as 8% while they were 6.6% in the original WRA.

The original WRA was carried out using WAsP. In order to account for the impact of

flow separation, WAsP and WindSim are employed in the present study. WindSim is

found to make closer estimations to the operational data by approximately 2-2.5%. This

63

result is consistently obtained for all the simulations and mainly caused by the wind flow

modelling of the WAsP which does not account for the orography induced turbulence.

Use of WAsP out of its suggested envelope contributes to the discrepancy to a limited

extent but cannot explain the discrepancy by itself.

The wind farm is subjected to considerable wake losses. Especially from the south-south

east direction where the adjacent wind farm is located, 15% of total wake losses occur in

this direction according to the WindPro power production assessment. Also, it was found

that the use of different wake models lead to substantially different results. In WindSim,

the Jensen and Ishihara wake model predicted 8.4-8.6% wake losses while the wake

losses using the Larsen model was only 4.5%. The original Jensen wake model is chosen

for the final simulations in the both software depending on the literature review and the

availability in both software.

The AEP data show big monthly and annual fluctuations, indicating, a high year-to-year

wind speed variability. Long term correlation and its potential influence on the AEP are

studied by analysing two different reanalysis datasets in the vicinity of wind farm as

long term references. Both of the long term references displayed quite decent and similar

correlation coefficients with the on-site time series. Nevertheless, the use of two

different long term references caused up to 5% difference in the AEP estimations.

WindSim calculated the Net AEP as 48.5GWh/y and 50.9GWh/y using CFSR-E and

MERRA data as long term reference respectively. The average value of three years

actual production data was 44.8GWh/y therefore it can be said that the use of CFSR-E

dataset as long term reference led closer results to the actual production data. The largest

deviations in the AEP estimations are encountered when the long term period or the long

term reference data is changed. It is concluded that the long term correlation might be a

factor accounting for a considerable part of the discrepancy. It is also concluded that

while choosing long term reference data, more parameters than the correlation

coefficient needs to be considered.

64

Further Research

According to Poulos 2015, underestimation of the curtailment losses and

large power curve under-performance are common reasons for the

underperformance of the wind farms. Curtailment losses are very case-

dependent and could not be investigated in this work due to the absence of

the related data. Also in this paper, power curve adjustment losses are

assessed using the typical ranges for the modern turbines. An investigation

for the contribution of these losses to the discrepancy might be beneficial.

An extensive research focusing solely on the long term correlation might

enable us to see the influence of this process better on the WRA for the case

of this study.

Even though the very similar correlation coefficients are obtained for both of

the long term references. The resulting AEP estimations were quite different.

To what extend the correlation coefficient is reliable and which other factors

might be considered for a reliable long term correlation?

Employing different wake models led to the different results. Especially the

results obtained using the Larsen and Jensen wake model are quite different.

The representativeness of the wake models for the actual losses occurring in

the studied wind farm might be investigated.

65

REFERENCES

Berge, E. et al., 2006. Wind in complex terrain. A comparison of WAsP and two CFD-

models, Available at:

http://www.windsim.com/documentation/papers_presentations/0602_ewec/ewec_b

erge.pdf.

Brower, M.C., 2012. WIND RESOURCE ASSESSMENT, A Practical Guide to

Developing a Wind Project, New Jersey: John Wiley and Sons, Ltd.

Campbell Scientific, 2016. Specifications of A100LK anemometers. Available at:

https://www.campbellsci.eu/a100lk [Accessed May 19, 2016].

Charhouni, N., Arbaoui, A. & Sallaou, M., 2015. Qualification of three analytical wake

models. 22ème Congrès Français de Mécanique. Available at:

http://documents.irevues.inist.fr/bitstream/handle/2042/57644/66892.pdf?sequence

=1.

CIRES, Advancing Reanalysis. Available at: http://reanalyses.org/ [Accessed May 5,

2016].

Coquilla, R. V., Obermeier, J. & White, B.R., 2007. Anemometer Calibration

Uncertainty. American Wind Energy Association, Windpower. Available at:

http://research.engineering.ucdavis.edu/wind/wp-

content/uploads/sites/17/2014/03/AWEA-2007-Final-Paper.pdf.

Diamond, K.E. & Crivella, E.J., 2011. Wind Turbine Wakes , Wake Effect Impacts , and

Wind Leases : Using Solar Access Laws As the Model for Capitalizing on Wind

Rights During the Evolution of Wind Policy Standards. Duke Environmental Law

& Policy Forum, 22:195, pp.195–244.

DNV, 2011. Wind Power Project Underperformance White Paper. Available at:

http://www.dnv.pl/Binaries/5. Wind Power Project Underperformance White

Paper_tcm144-517779.pdf.

Gallagher, J., 2014. an Assessment of the Discrepancy Between Operational Assessment

and Wind Resource Assessment for a Wind Farm in Ireland.

GH and Partners, 2009. GH WindFarmer: Theory Manual,

Liléo, S. et al., 2013. Long-term correction of wind measurements, State-of-the-art and

guidelines for the future work, Available at:

http://www.elforsk.se/Global/Vindforsk/Rapporter

VFIII/13_18_report_V377_longtermcorrections.pdf.

Lira, A.G. et al., 2014. Uncertainties in the estimate of wind energy production. , p.12.

Mancebo, C.D., 2014. Comparison Study for Wind Resource Assesment in Complex.

Uppsala University.

Meissner, C., 2010. WindSim, Getting started. WindSim 5. 5th Edition, Available at:

http://windsim.com/ws5_help/ws_docs/Getting Started.pdf.

Mortensen, N.G., 2011. Planning and Development of Wind Farms : Wind Resource

Assessment and Siting,

Mortensen, N.G., Steen, R.O. & Morten, N., 2011. Wind Atlas Analysis and Application

Program: WAsP 10 Help Facility, Available at:

http://orbit.dtu.dk/files/116352660/WAsP_10_Help_Facility.pdf.

66

NASA, 2012. MERRA: MODERN ERA-RETROSPECTIVE ANALYSIS FOR

RESEARCH AND APPLICATIONS. Global Modeling and Assimilation Office.

Available at: http://gmao.gsfc.nasa.gov/research/merra/intro.php [Accessed May 5,

2016].

NCAR, 2016. Climate Forecast System Reanalysis (CFSR). Available at:

https://climatedataguide.ucar.edu/climate-data/climate-forecast-system-reanalysis-

cfsr [Accessed May 5, 2016].

Neustadter, H.E. & Spera, D.A., 1985. Method for Evaluating Wind Turbine Wake

Effects on Wind Farm Performance. J. Sol. Energy Eng 107(3),, pp.240–243.

Available at:

http://solarenergyengineering.asmedigitalcollection.asme.org/article.aspx?articleid=

1454440.

Nilsson, K., 2010. Estimation of wind energy production in relation to orographic

complexity-A reliability study of two conventional computer software. Gotland

University.

Poulos, G.S., 2015. The Most Important Factors in Wind Farm Under-Performance. In

AWEA Wind Resource & Project Energy Assessment Seminar 2015. V-Bar, LLC.

Renkema, D.J., 2007. Validation of wind turbine wake models Using wind farm data and

wind tunnel measurments. TU Delft. Available at:

http://lr.home.tudelft.nl/fileadmin/Faculteit/LR/Organisatie/Afdelingen_en_Leersto

elen/Afdeling_AEWE/Wind_Energy/Education/Masters_Projects/Finished_Master

_projects/doc/Douwe_Renkema_r.pdf.

Simisiroglou, N., 2012. WIND RESOURCE ASSESSMENT COMPARISON ON A

COMPLEX TERRAIN EMPLOYING WINDPRO AND WINDSIM. Gotland

University.

Sørensen, T., Thøgersen, M.L. & Nielsen, P., 2008. Adapting and calibration of existing

wake models to meet the conditions inside offshore wind farms, EMD International

A/S. Available at: http://www.emd.dk/files/PSO projekt 5899.pdf.

Teneler, G. (Högskolan på G., 2011. Wind Flow Analysis on a Complex Terrain: a

reliability study of a CFD tool on forested area including effects of forest module.

Gotland University. Available at: http://hgo.diva-

portal.org/smash/record.jsf?pid=diva2:458063.

Thøgersen, M.L. et al., 2007. Measure-correlate-predict methods: case studies and

software implementation. European Wind Energy Conference & Exhibition., p.10.

Available at: http://emd.dk/files/windpro/Thoegersen_MCP_EWEC_2007.pdf.

Timander, T. & Westerlund, J., 2012. Reliability of wind farm design tools in complex

terrain: A comparative study of commercial software. KTH School of Industrial

Engineering and Management.

Tong, W. et al., 2012. Impact of Different Wake Models On the Estimation of Wind

Farm Power Generation. 12th AIAA Aviation Technology, Integration, and

Operations (ATIO) Conference and 14th AIAA/ISSM, pp.1–13. Available at:

https://messac.expressions.syr.edu/wp-

content/uploads/2012/10/Messac_2012_MAO-WFO2.pdf.

Troen, I. & Lundtang Petersen, E., 1989. European wind atlas, Available at:

67

http://orbit.dtu.dk/en/publications/european-wind-atlas(335e86f2-6d21-4191-8304-

0b0a105089be).html.

Wallbank, T., 2008. WindSim Validation Study: CFD validation in Complex terrain,

Available at:

http://web.windsim.com/documentation/papers_presentations/thesis/080512trw

WindSim Write Up - Validation

study.pdf\nhttp://www.windsim.com/documentation/papers_presentations/thesis/08

0512trw WindSim Write Up - Validation study.pdf.

68

APPENDIX A. Windographer Tower Shading Flagging Tool

APPENDIX B. Weighted Mean of Sector Wise Correlations

Figure 16: Weighted mean of sector wise correlation between M1 and MERRA

Figure 15: Tower Shading Flagging in Windographer

69



Figure 19: Weighted mean of sector wise correlation between M1 and CFSR-E

70



APPENDIX C. Convergence Table

Table 17: Convergence table Source: WindSim

Sectors Simulation

time Iterations Status Sectors

Simulation time

Iterations Status

000 00:14:38 76 Converged 180 00:15:21 80 Converged






71

APPENDIX D. Climatology characteristics including average wind speed (m/s) for

all sectors, Weibull shape (k) and scale (A) parameters for all sectors

Figure 22: Climatology characteristics for M1/MERRA



72

Figure 25: Climatology characteristics for M1/CFSR-E



investigation of potential reasons to account for …949713/fulltext01.pdf · construction, a wind...

Documents